Beruflich Dokumente
Kultur Dokumente
Note:Thisisnotaparticularlyinterestingtopicforanyoneotherthanthosewhoareinvolvedinsimulation.Soifyouhavedifficultywiththisissue,you
maysafelystopreadingafterpage5withoutfeelingguilty.
HilbertTransform,AnalyticSignalandtheComplexEnvelope
InDigitalSignalProcessingweoftenneedtolookatrelationshipsbetweenrealandimaginarypartsofacomplex
signal.TheserelationshipsaregenerallydescribedbyHilberttransforms.Hilberttransformnotonlyhelpsusrelate
theIandQcomponentsbutitisalsousedtocreateaspecialclassofcausalsignalscalledanalyticwhichare
especiallyimportantinsimulation.Theanalyticsignalshelpustorepresentbandpasssignalsascomplexsignals
whichhavespeciallyattractivepropertiesforsignalprocessing.
HilbertTransformisnotaparticularlycomplexconceptandcanbemuchbetterunderstoodifwetakeanintuitive
approachfirstbeforedelvingintoitsformulawhichisrelatedtoconvolutionandishardtograsp.Thefollowing
diagramthatisoftenseenintextbooksdescribingmodulationgivesusaclueastowhataHilbertTransformdoes.
Figure1RoleofHilbertTransforminmodulation
TheroleofHilberttransformaswecanguesshereistotakethecarrierwhichisacosinewaveandcreateasine
waveoutofit.SoletstakeacloserlookatacosinewavetoseehowthisisdonebytheHilberttransformer.Figure
2ashowstheamplitudeandthephasespectrumofacosinewave.NowrecallthattheFourierSeriesiswrittenas
where
and
Cosinewavehasnosinespectralcontent,soBn iszero.Thephasecalculatedis90forbothpositiveand
negativefrequencyfromaboveformula.Thewavehastwospectralcomponentseachofmagnitude1/2A,both
positiveandlyingintherealplane.(therealplaneisdescribedasthatpassingvertically(RVplane)andthe
Imaginaryplaneasonehorizontally(RIplane)throughtheImaginaryaxis)
Figure2bshowsthesametwospectrumsforasinewave.Thesinewavephaseisnotsymmetricbecausethe
amplitudespectrumisnotsymmetric.ThequantityAn iszeroandBn haseitherapositiveornegativevalue.The
phaseis+90forthepositivefrequencyand90forthenegativefrequency.
Nowwewishtoconvertthecosinewavetoasinewave.Therearetwowaysofdoingthat,oneintimedomainand
theotherinfrequencydomain.
HilbertTransforminFrequencyDomain
NowcompareFigure2aand2b,inparticularthespectralamplitudes.Thecosinespectralamplitudesareboth
positiveandlieintherealplane.ThesinewavehasspectralcomponentsthatlieintheImaginaryplaneandareof
oppositesign.
Toturncosineintosine,asshowninFigure3below,weneedtorotatethenegativefrequencycomponentofthe
cosineby+90andthepositivefrequencycomponentby90.Wewillneedtorotatethe+Qphasorby90orin
otherwordsmultiplyitbyj.WealsoneedtorotatetheQphasorby+90ormultiplyitbyj.
Figure3Rotatingphasorstocreateasinewaveoutofacosine
WecandescribethistransformationprocesscalledtheHilbertTransformasfollows:
Allnegativefrequenciesofasignalgeta+90phaseshiftandallpositivefrequenciesgeta90phaseshift.
Ifweputacosinewavethroughthistransformer,wegetasinewave.Thisphaserotationprocessistrueforall
signalsputthroughtheHilberttransformandnotjustthecosine.
Foranysignalg(t),itsHilbertTransformhasthefollowingproperty
(PuttingalittlehatoverthecapitalletterrepresentingthetimedomainsignalisthetypicalwayaHilbertTransformiswritten.)
AsinewavethroughaHilbertTransformerwillcomeoutasanegativecosine.Anegativecosinewillcomeouta
negativesinewaveandonemoretransformationwillreturnittotheoriginalcosinewave,eachtimeitsphasebeing
changedby90.
ForthisreasonHilberttransformisalsocalledaquadraturefilter.Wecandrawthisfilterasshownbelowin
Figure4.
Figure4HilbertTransformshiftsthephaseofpositivefrequenciesby90andnegativefrequenciesby+90.
SoherearetwothingswecansayabouttheHilbertTransform.
1.Itisapeculiarsortoffilterthatchangesthephaseofthespectralcomponentsdependingonthesignoftheir
frequency.
2.Itonlyeffectsthephaseofthesignal.Ithasnoeffectontheamplitudeatall.
HilberttransforminTimeDomain
Nowlookatthesignalintimedomain.Givenasignalg(t),HilbertTransformofthissignalisdefinedas
(1)
AnotherwaytowritethisdefinitionistorecognizethatHilbertTransformisalsotheconvolutionoffunction
1/ptwiththesignalg(t).Sowecanwritetheaboveequationas
(2)
AchievingaHilbertTransformintimedomainmeansconvolvingthesignalwiththefunction1/pt.Whythe
function1/pt,whatisitssignificance?LetslookattheFourierTransformofthisfunction.Whatdoesthattellus?
GiveninEq3,thetransformlooksalotliketheHilberttransformwetalkedaboutbefore.
(3)
ThetermsgninEq3above,calledsignumissimplerthanitseems.Hereisthewaywecouldhavewritten
itwhichwouldhavebeenmoreunderstandable.
(4)
InFigure5weshowthesignumfunctionanditsdecompositionintotwofamiliarfunctions.
Figure5SignumFunctiondecomposedintoaunitfunctionandaconstant
Forshortcut,writingsgnisusefulbutitisbetterifitisunderstoodasasumoftheabovetwomuchsimpler
functions.(Wewillusethisrelationshiplater.)
(5)
Weseein6figurethatalthough1/ptisarealfunction,ishasaFouriertransformthatliesstrictlyinthe
imaginaryplane.DoyourecallwhatthismeansintermsofFourierSeriescoefficients?Whatdoesittellusabouta
functionifithasnorealcomponentsinitsFouriertransform?Itsaysthatthisfunctioncanberepresentedcompletely
byasumofsinewaves.Ithasnocosinecomponentatall.
InFigure7,weseeafunctioncomposedofasumof50sinewaves.Weseethesimilarityofthisfunction
withthatof1/pt.Nowyoucanseethatalthoughthefunction1/ptlooksnothingatallasinusoid,wecanstill
approximateitwithasumofsinusoids.
Thefunctionf(t)=1/ptgivesusaspectrumthatexplainstheHilbertTransformintimedomain,albeitthis
wayoflookingattheHilbertTransformisindeedveryhardtograsp.
WelimitourdiscussionofHilberttransformtoFrequencydomainduetothisdifficulty.
Figure7Approximatingfunctionf(t)=1/ptwithasumof50sinewave
WecanaddthefollowingtoourlistofobservationsabouttheHilbertTransform.
3.ThesignalanditsHilbertTransformareorthogonal.Thisisbecausebyrotatingthesignal90wehavenow
madeitorthogonaltotheoriginalsignal,thatbeingthedefinitionoforthogonality.
4.ThesignalanditsHilbertTransformhaveidenticalenergybecausephaseshiftdonotchangetheenergyofthe
signalonlyamplitudechangescandothat.
AnalyticSignal
HilbertTransformhasotherinterestingproperties.OneofthesecomesinhandyintheformulationofanAnalytic
signal.AnalyticsignalsareusedinDoubleandSinglesidebandprocessing(aboutSSBandDSBlater)aswellas
increatingtheIandQcomponentsofarealsignal.
Ananalyticsignalisdefinedasfollows.
(6)
AnanalyticsignalisacomplexsignalcreatedbytakingasignalandthenaddinginquadratureitsHilbert
Transform.Itisalsocalledthepreenvelopeoftherealsignal.
Sowhatistheanalyticsignalofacosine?
Substitutecoswtforg(t)inEq6,knowingthatitsHilberttransformisasine,weget
Theanalyticfunctionofacosineisthenowfamiliarphasororthecomplexexponential,ejwt.
Whatistheanalyticsignalofasine?
Nowsubstitutesinwtforg(t)inEq6,knowingthatitsHilberttransformisacos,wegetonceagainacomplex
exponential.
Doyourememberwhatthespectrumofacomplexexponentiallookslike?Toremindyou,Irepeatherethefigure
fromTutorial6.
Figure8Fouriertransformofacomplexexponential
Wecanseefromthefigureabove,thatwhereasthespectrumofasineandcosinespansboththenegativeand
positivefrequencies,thespectrumoftheanalyticsignal,inthiscasethecomplexexponential,isinfactpresentonly
inthepositivedomain.Thisistrueforbothsineandcosineandinfactforallrealsignals.
Restatingtheresults:theAnalyticsignalforbothandsineandcosineisthecomplexexponential.Even
thoughbothsineandcosinehaveatwosidedspectrumasweseeinfiguresabove,thecomplexexponential
whichistheanalyticsignalofasinusoidhasaonesidedspectrum.
Wecangeneralizefromthis:Ananalyticsignal(composedofarealsignalanditsHilberttransform)hasaspectrum
thatexistsonlyinthepositivefrequencydomain.
Letstakeatalookattheanalyticsignalagain.
(7)
Theconjugateofthissignalisalsoausefulquantity.
(8)
Thissignalhascomponentsonlyinthenegativefrequenciesandcanbeusedtoseparateoutthelowersidebands.
Nowbacktotheanalyticsignal.LetsextendourunderstandingbytakingFourierTransformofbothsidesofEq7.
Weget
(9)
ThefirsttermistheFouriertransformofthesignalg(t),andthesecondtermistheinverseHilbertTransform.We
canrewritebyuseofproperty Eq9as
(10)
Onemoresimplificationgivesus
(11)
Thisisaveryimportantresultandisapplicabletobothlowpassandmodulatedsignals.Formodulatedorbandpass
signals,itsneteffectistotranslatethesignaldowntobaseband,doublethespectralmagnitudesandthenchopoff
allnegativecomponents.
ComplexEnvelope
Wecannowdefineanewquantitybasedontheanalyticsignal,calledtheComplexEnvelope.TheComplex
Envelopeisdefinedas
Thepart iscalledtheComplexEnvelopeofthesignalg(t).
LetsrewriteitandtakeitsFourierTransform.
WenowseeclearlythattheComplexEnvelopeisjustthefrequencyshiftedversionoftheanalyticsignal.
Recognizingthatmultiplicationwiththecomplexexponentialintimedomainresultsinfrequencyshiftinthe
Frequencydomain,usingtheFourierTransformresultsfortheanalyticsignalabove,weget
(12)
Sohereiswhatwehavebeentryingtogetatallthistime.ThisresultsaysthattheFourierTransformoftheanalytic
signalisjusttheonesidedspectrum.Thecarriersignaldropsoutentirelyandthespectrumisnolongersymmetrical.
Thispropertyisveryvaluableinsimulation.Wenolongerhavetodosimulationatcarrierfrequenciesbutonlyat
thehighestfrequencyofthebasebandsignal.Theprocessappliesequallytoothertransformationsuchasfiltersetc.
whicharealsodownshifted.Itevenworkswhennonlinearitiesarepresentinthechannelandresultinadditional
frequencies.
Thereareotherusesofcomplexrepresentationwhichwewilldiscussasweexplorethesetopicshoweveritsmain
useisinsimulation.
__________________________________
Example
Letsdoanexample.Hereisarealbasebandsignal.
(Ihaveleftoutthefactor2pforpurposesofsimplification)
Figure9ABasebandSignal
Thespectrumofthissignalisshownbelow,bothitsindividualspectralamplitudesanditsmagnitudespectrum.The
magnitudespectrumshowsonespectralcomponentofmagnitude2atf=2and2andananotheroneofmagnitude
3atf=3and3.
Figure10aSpectralamplitudesFigure10bTheMagnitudeSpectrum
Nowletsmultiplyitwithacarriersignalofcos(100t)tomodulateitandtocreateabandpasssignal,
Figure11Themodulatedsignalanditsenvelope
LetstaketheHilbertTransformofthissignal.Butbeforewedothatweneedtosimplifytheabovesoweonlyhave
sinusoidsandnottheirproducts.ThisstepwillmakeiteasytocomputetheHilbertTransform.Byusingthese
trigonometricrelationships,
werewritetheabovesignalas
NowwetaketheHilbertTransformofeachtermandget
NowcreatetheanalyticsignalbyaddingtheoriginalsignalanditsHilbertTransform.
Letsonceagainrearrangethetermsintheabovesignal
RecognizingthateachpairoftermsistheEulersrepresentationofasinusoid,wecannowrewritetheanalytic
signalas
Butwaitaminute,isntthistheoriginalsignalandthecarrierwritteninthecomplexexponential?Sowhyallthe
calculationsjusttogettheoriginalsignalback?
NowletstaketheFourierTransformoftheanalyticsignalandthecomplexenvelopewehavecomputedto
showtherealadvantageofthecomplexenveloperepresentationofsignals.
SpectrumoftheComplexEnvelopeSpectrumoftheAnalyticSignal
Figure12TheMagnitudeSpectrumoftheComplexEnvelopevs.TheAnalyticSignal
Althoughthiswasapassbandsignal,weseethatitscomplexenvelopespectrumiscenteredaroundzeroandnotthe
carrierfrequency.Alsothespectralcomponentsaredoublethoseinfigure10bandtheyareonlyonthepositive
side.IfyouthinktheresultlookssuspiciouslylikeaonesidedFouriertransform,thenyouwouldberight.
WedoallthisbecauseofsomethingNyquistsaid.Hesaidthatinordertoproperlyreconstructasignal,anysignal,
basebandorpassband,needstobesampledatleasttwotimesitshighestspectralfrequency.Thatrequiresthatwe
sampleatfrequencyof200.
Butwejustshowedthatifwetakeamodulatedsignalandgothroughallthismathandcreateananalyticsignal
(whichbythewaydoesnotrequireanyknowledgeoftheoriginalsignal)wecanseparatetheinformationsignalthe
basebandsignals(t))fromthecarrier.Wedothisbydividingtheanalyticsignalbythecarrier.Nowallwehaveleft
isthebasebandsignal.Allprocessingcanbedoneatasamplingfrequencywhichis6(twotimesthemaximum
frequencyof3)insteadof200.
ThepointhereisthatthismathematicalconcepthelpusgetaroundthesignalprocessingrequirementsbyNyquist
forsamplingofbandpasssystems.
Thecomplexenvelopeisusefulprimarilyforpassbandsignals.Inalowpasssignalthecomplexenvelopeofthe
signalisthesignalitself.Butinpassbandsignal,thecomplexenveloperepresentationallowsustoeasilyseparate
outthecarrier.
Takealookatthecomplexenvelopeagainforthissignal
theanalyticsignal
thecomplexenvelope
Weseetheadvantageofthisformrightaway.Thecomplexenvelopeisjustthelowpasspartoftheanalyticsignal.
Theanalyticsignallowpasssignalhasbeenmultipliedbythecomplexexponentialatthecarrierfrequency.The
Fouriertransformofthisrepresentationwillleadtothesignaltranslatedbackdownthebaseband(anddoubledwith
nonegativefrequencycomponents)makingitpossibletogetaroundtheNyquistsamplingrequirementandreduce
computationalload.
Copyright1999CharanLangton,AllRightsReserved
mntcastle@earthlink.net