devices in Simulink
EYAD ARABI & SADIQ ALI
EX021/2008
III
IV
Abstract
Simulation of wireless systems has become a key issue in analyzing, optimizing and
designing wireless systems. In this thesis, modeling RF front end devices in
Simulink is investigated. The capabilities of Simulink and RF Blockset are tested.
Their different behavioral models for nonlinearity, noise, phase noise and mismatch
are analyzed. A model for the power amplifier that takes into account memory effects
is implemented in Simulink to extend the RF Blockset to model wideband
applications such as WCDMA. This model for the power amplifier implements a
memorypolynomial model. Memorypolynomials prove to be both accurate and easy
to implement.
Keywords: RF front end modeling, behavioral modeling, power amplifier modeling,
memorypolynomial, Simulink, RF Blockset.
Acknowledgment
TheauthorswishtothankProfessorMatsVibergforhissmoothandeffective
supervision, Professor Thomas Ericsson for his invaluable directions and
comments and Ali Soltani who has provided the measurement data for the
poweramplifierusedinthiswork.
VI
VII
Contents
ABSTRACT
ACKNOWLEDGMENT
1
INTRODUCTION
V
VI
1
1.1
RFfrontendsystems
1.1.1
RFTransmitters
1.1.2
RFReceivers
1
2
3
1.2
Typesofmodeling
1.2.1
Physicalmodeling
1.2.2
Behavioralmodeling
3
3
3
1.3
BasebandandPassbandmodeling
1.4
Harmonicsandintermodulationproductsinnonlineardevices
Introduction
7
7
2.2
Poweramplifiermodeling
2.2.1
AMAMandAMPMModeling
2.2.2
Memorylessandquasimemorylessnonlinearmodels
2.2.3
Nonlinearmodelswithmemory
2.2.4
Comparisonsbetweennonlinearmodelswithmemory
8
9
10
13
20
2.3
MemoryPolynomialModel
2.3.1
Modelsimplementation
2.3.2
ModelsIdentification
21
21
23
LOWNOISEAMPLIFIER
3.1
Introduction
3.2
LNAsModelinRFBlockset
3.2.1
InputportandOutputportBlocks
3.2.2
MismatchseffectsmodelinginRFBlockset
3.2.3
Scatteringmatrix
3.2.4
ModelingofNoiseinRFBlockset
3.2.5
ModelingofNonlinearityofLNA
4
VII
MIXERMODELLING
27
27
28
29
34
34
35
37
41
4.1
Introduction
41
4.2
ModelingofMixersinRFBlockset
42
4.3
ModelingofPhaseNoise
4.3.1
PhasenoisemodelinginRFBlockset
44
44
ANALYSISOFMEMORYPOLYNOMIALMODEL
5.1
Introduction
5.2
Modelsanalysis
5.2.1
AM/AMandAM/PMPlots
5.2.2
Multitonetest
5.2.3
ResponsetoWCDMAsignal
6
CONCLUSIONSANDFUTUREWORK
47
47
47
47
51
52
55
APPENDIXA
57
APPENDIXB
63
REFERENCELIST
67
VIII
Chapter 1 Introduction
Introduction
1.1
RFfrontendsystems
An RF front end system refers to the analog front end of the wireless
communication system. The digital baseband signals cannot be transmitted
directly through wireless channels due to properties of the electromagnetic
waves.Therefore,thesesignalsmustbeconvertedtoanalog,upconvertedto
higherfrequencies,andtransmittedthroughthechannel.Thereceivedsignals
are down converted to the baseband frequency then converted to digital
again. An overall wireless communication system is shown in Fig. 1.1.
Processes done to the analog signal in the RF front end includes filtering,
amplification, and mixing. These processes are imperfect and add various
impairments to the received signal, in this thesis the modeling of major
impairments added by each RF component is investigated, and an overall
modelforthereceiverandthetransmitterareimplementedinSimulink.
Fig.1.1Aschematicofanoverallwirelesssystemstructure.Antennasaresometimes
includedwithRFfrondend.
1.1.1
RFTransmitters
Fig.1.2. Shows a block diagram for a typical RF transmitter, only the power
amplifier and the mixer are included in the modeling, because these
components add the most serious impairments to the transmitter. Other
components that do not add serious impairments, like filters, are ignored in
this thesis. The mixer introduces phase noise, spurious frequencies and
nonlinearity. The power amplifier introduces nonlinearity. The models for
theseblocksareillustratedinsubsequentchapters.
Fig.1.2ASchematicofanRFtransmitter.Onlycomponentsthatintroduceeffective
impairmentsareincluded.
Chapter 1 Introduction
1.1.2
RFReceivers
Fig.1.3. shows a block diagram for an RF receiver. Only the low noise
amplifier (LNA) and the mixer are included in the model. The LNA
introduces noise and nonlinearity. The mixer introduces phase noise and
nonlinearity.
Fig.1.3AschematicofanRFreceiver.Onlyblocksthatintroduceeffective
impairmentsareincluded.
1.2
Typesofmodeling
1.2.1
Physicalmodeling
Aphysicalmodelrequiresknowledgeoftheelementsthatcomprisethereal
system, their constitutive relations, and the theoretical rules describing their
interactions. These types of modeling are appropriate for circuit level
simulationandcanbeveryaccurate[17].
1.2.2
Behavioralmodeling
A behavioral model, also called black box model, does not require prior
knowledgeofthephysicalsystems.Itsinternalstructurereliesonlyoninput
andoutmeasurements.Theparametersofbehavioralmodelsareindentified
frominputandoutputmeasurementdata.Thereforemeasurementtechniques
andthequalityofdataaffecttheaccuracyofthesemodels[17].Allthemodels
presentedinthisthesisarebehavioral.
3
1.3
BasebandandPassbandmodeling
Apassbandsignalcanberepresentedbythefollowingequation[4]:
cos
(1.1)
where
isthecentercarrierfrequency
istheamplitudeofthesignal
isthemodulatedphase
Thispassbandsignalcanberewrittenas
cos
where
cos
sin
sin
and
(1.2)
quadraturecomponentsrespectively.Equation(1.2)canberewrittenas
cos
cos
sin
.
(1.3a)
(1.3b)
sin
(1.3c)
where
isthecomplexbasebandsignal,whichcanberepresentedby
(1.4a)
(1.4b)
Simulationscanbedoneeitherinpassbandorcomplexbaseband.Passband
simulations are simpler and more accurate. However, they consume more
resources and simulation time. To show this, assume a signal of center
4
Chapter 1 Introduction
frequency andanoperationalbandwidth
time for this signal in Passband,
contrast, baseband simulations require
.Tosimulateonesecondofreal
1.4
,where isaninteger
greaterthanone.Thesefrequenciesarecalledharmonics.
If a two tone signal with frequencies
and
is applied to a nonlinear
where
and are two integers greater than zero. Theses frequencies are called
intermodulation products. The order of a given intermodulation product is
defined as  
devicesisthe3rdorderintermodulationproduct,becauseitisusuallycloseto
thedesired frequencyandcannotbecompletelyfilteredout.Thisproductis
3.2. It can be shown that the slope of the linear gain for input and output
powersindBsisunity,likewisetheslopeofthethirdgainofthethirdorder
intermodulation component is 3 [7], the point where the third order line
intersectswiththelineargainlineisthethirdorderinterceptpoint.
Fig.3.2.Anillustrationofthefirstandthirdorderinterceptspoints.
Anotherfigureofmerittocharacterizenonlinearityisthe1dBcompression
point.Foranonlineardevice,the1dBcompressionpointisdefinedasthe
pointwherethedifferencebetweenthedevicesoutputandthelinearoutput
isexactly1dB.The1dBcompressionpointistypically12to15dBlessthan
the3rdorderinterceptpointassumingtheyarereferencedatthesamepoint.
2.1
Introduction
bandapplicationsthemostimportantaspecttomodelisthedevicenonlinearity
thattheamplifierintroducestothesystem.
Thetarget applicationforthisworkisWCDMA.Animportantcharacteristicof
thisschemeisthelargebandwidth.Largebandwidthgenerallyleadstohigher
frequency of the signal envelop, this feature will bring another important
phenomena to the surface that is memory effect. Almost every device has a
memory temporal dynamics or response delay, these effects are due to the
biasing of the circuit and the capacitance or inductance built in the device. For
wide band applications, memory effects cannot be ignored. Therefore, a model
that takes into account memory effects must be used to accurately model the
poweramplifiersdistortion.
InSection2.2anoverviewofthemostcommonlyconsideredmodelsforpower
amplifiersarepresented,acomparisonbetweenthesemodelsisalsodiscussedto
selectthemostsuitablemodel.InSection2.3theimplementationoftheselected
modelinSimulinkisexplainedindetail.
2.2
Poweramplifiermodeling
Thepoweramplifiersnonlinearitybroadenstheinputsignalsbandwidth.This
is known as Spectral regrowth and is undesired. Spectral regrowth causes
interferencewithadjacentchannelsandincreasestheprobabilityofviolationsof
the outofband emission requirements mandated by regulatory bodies. It also
causesdistortionswithinthesignalbandwidth,whichaffectsthebiterrorrateat
thereceiver.Mostrecenttransmissionschemes,suchasWidebandCodeDivision
multiple access (WCDMA) or Orthogonal Frequency Division Multiplexing
(OFDM), are especially vulnerable to the nonlinear distortions due to high
fluctuationsintheirpowerlevels.
2.2.1
AMAMandAMPMModeling
The AMAM conversion for a nonlinear system is the relation between the
amplitude of the systems output and the amplitude of the systems input. The
AMPM conversion for a nonlinear system is the relation between the phase
changeofthesystemsinputandoutput,andtheamplitudeoftheinputsignal.
ThisisshowninFig.2.1.
Assumingthepassbandinputsignal(1.1),theoutputoftheAMAMandAMPM
canbewrittenas
model
cos
(2.1)
where
istheamplitudenonlinearityorAMAMconversions
isthephasenonlinearityorAMPMconversions
Fig.2.1.Amplitudephasenonlinearmodelstructureforacomplexbasebandinputandoutput
signals.
InthefollowingSectionsdifferentmodelsforpoweramplifiersarepresented.
2.2.2
Memorylessandquasimemorylessnonlinearmodels
Inmemoryless(static)poweramplifiermodelstheoutputsignalisanonlinear
functionofthecurrentinputsignalonlyand previousvaluesofthesignalhave
noeffectontheoutputofthemodel.MemorylessmodelsonlyconsiderAMAM
conversions, and assume no phase change. On the other hand Quasimemory
less models take into account both amplitude and phase distortions. Therefore,
they are represented by the amplifier AM/AM as well as AM/PM transfer
functions.
Static models give reasonable accuracy for applications with a narrowband
frequency spectrum or when memory effects are not important. Quasi
memoryless models have better accuracy for narrowband applications. In the
following subsections some memoryless and quasimemoryless models are
discussed.MostofthesearealreadyimplementedinMatlabSimulink.
2.2.2.1 Polynomialmodel
Thepolynomialmemorylesspoweramplifiermodelcanberepresentedinbase
bandbythefollowingequation
(2.2)
where
istheinputcomplexbasebandsignal.
isoutputcomplexbasebandsignal.
arerealvaluedcoefficients.
10
memorylesspolynomialmodelforthepoweramplifier[2].
2.2.2.2 TheRappModel
TheRappmodelusesthreeparameters,andmodelsamplitudedistortionbutno
phasedistortion.ThegeneralexpressionoftheAMAMconversionsisasfollows
(2.3)
where
isaparameterthatsetstheoutputsaturationlevel.
is a parameter that sets the smoothness of the transition from linear to
saturationstates,thesmallerSthesmootherthetransition.
Thetechniqueofthismodelisquitesimple.Itassumeslinearperformanceuntil
the saturation point is approached. When the saturation point is approached, a
transitiontowardsaconstantsaturatedoutputisapplied[18].
2.2.2.3 TheSalehmodel
TheSalehmodelisaquasimemorylessmodel.Itusesfourparameterstofitthe
model to measurement data. Its AMAM and AMPM conversion functions are
describedbythefollowingequations
(2.4a)
(2.4b)
where
arethemodelsparameters,[13].
11
2.2.2.4 TheGhorbanimodel
The Ghorbani model uses eight parameters to fit the model to measurement
data,thismodelisquasimemoryless,anditsAMAMandAMPMconversions
functionsaredescribedbythefollowingequations
(2.5a)
(2.5b)
where
,
measurementdatabymeansofcurvefitting[12].
2.2.2.5 TheHyperbolictangentmodel
TheHyperbolictangentmodelisquasimemoryless.Ithasfiveparametersthese
are,IIP3,lineargainG,upperpowerlimit
phase gain
,lowerpowerlimit andthelinear
tanh
3
3
(2.6a)
where
3 isthethirdorderinterceptpointreferredtotheinput.
G isthelineargain.
The AMPM conversion is linear and is specified by the slope of the AMPM
conversion
indegree/dB.Thislinearityisboundedbytwoparameterswhich
12
are
and
.Ifthepowermagnitudeoftheinputislessthan
thennophase
distortionisadded,andifthepower magnitudeoftheinputisgreaterthan
then a constant phase shift of
bythefollowingequation



(2.6b)
where
istheslopeoftheAMPMlinearity.
and aretheupperandlowerpowerlimitparametersrespectively.
2.2.3
Nonlinearmodelswithmemory
Inrealitytheoutputofthepoweramplifierdependsonpreviousinputsaswell
asthecurrentinputoftheamplifier.Thisphenomenoniscalledmemoryeffect,
orsimplytemporaldynamics.Thesememoryeffectsare dueto thermaleffects,
andlongtimeconstantsinDCbiascircuits.Itcanbeobservedasasymmetriesin
lower and upper sidebands, and bandwidth dependent variations in the
magnitudeofintermodulationproducts.Forhigherbandwidthapplications,e.g.
WCDMA, the memory effects becomes severe, and cannot be ignored. Hence,
memorylessandquasimemorylessmodelsarenotaccurateenough.Therefore,
amodelwhichconsidersmemoryeffectsshouldbeusedforsuchapplications.
In the following Sections some of the most common models with memory are
presented.
13
2.2.3.1 Volterraseries
The Volterra model can be used to describe any nonlinear stable system with
fadingmemory,withanarbitrarysmallerror.However,itsmaindisadvantages
arethedramaticincreaseinthenumberofparameterswithrespecttononlinear
order and memory length, which causes drastic increase of complexity in the
identificationofparameters.Thisisthereasonwhyitishighlyunpracticaltouse
volterraseriesforsystemswithhighnonlinearordersandmemorylengths.This
modelcanbeexpressedmathematicallyasfollows[2]
2
2
1, 2 . . , 2
(2.7)
FromtheequationaboveitisclearthatthenumberofcoefficientsoftheVolterra
series increases exponentially as the memory length and the nonlinear order
increase.
2.2.3.2 WienerandHammersteinmodels
AsmentionedaboveVolterraseriesisunpracticalformodelingpoweramplifiers
inrealtimeapplications.Thisreasonmotivatedresearcherstoinvestigatespecial
cases of Volterra series. The Wiener model, the Hammerstein model, and the
WienerHammerstein model are included in the category of special cases of
Volterraseriesformodelingnonlinearpoweramplifiers.
TheWienermodelconsistsofalineartimeinvariant(LTI)systemfollowedbya
memorylessnonlinearityasillustratedinFig.2.2.
14
Fig.2.2:WienerModel
The model is given by the following mathematical formulas. The output of the
LTIsystemisgivenas
(2.8a)
Theoutputofthestaticnonlinear(NL)blockisgivenas
2
(2.8b)
Inserting(2.8a)into(2.8b)gives
2
1
1
2
(2.8c)
Fig.2.3.Hammersteinmodel
15
Themodelisrepresentedbythefollowingequations.
(2.9a)
(2.9b)

1
(2.9c)
TheWienerHammerstein(WH)modelconsistsofanLTIsystemfollowedbya
memorylessnonlinearity,whichisinturnfollowedbyanotherLTIsystem.The
WienerHammersteinmodelisshowninFig.2.4
Fig.2.4:WienerHammersteinModel
Theoutputismodeledby[2]
1
1 ,
(2.10a)
1 0
2
(2.10b)
16
2
1
1 0
2
0
2
2
2 0
2 0
(2.10c)
where
and
areparameterstospecifymemorydepthsofLTI.
2.2.3.3 TheParallelHammersteinmodel
The ParallelHammerstein is an extension of the standard Hammerstein model.
ThemodelisillustratedinFig.2.5.Thesysteminthiscaseismodeledby[2][3]
2
(2.11a)
(2.11b)
where
istheinputcomplexbasebandsignal.
istheoutputcomplexbasebandsignal.
isthetransferfunctionofthefilterforthekthpolynomialcontribution.
,
arecomplexvaluedparameters.
isthememorydepth.
istheorderofthepolynomial.
The main difference between the ParallelHammerstein and the standard
HammersteinmodelsisthatintheParallelHammersteinmodel,differentstatic
nonlinear orders are filtered with different LTI systems. For example, the first
17
termofthepolynomialisfilteredwith
  isfilteredwith
, and the2ndoddpowertermi.e.
andsoon.
Fig.2.5BlockdiagramfortheParallelHammerstein
2.2.3.4 TheMemoryPolynomialModel
Thememorypolynomialmodel,[1]consistsofseveraldelaytapsandnonlinear
staticfunctions.ThismodelisatruncationofthegeneralVolterraseries,which
consistsofonlythediagonaltermsintheVolterrakernels.Thus,thenumberof
parameters is significantly reduced compared to general Volterra series. The
modelisshowninFig.2.6
18
Fig.2.6.Thememorypolynomialmodel.
2
0
1,
2
(2.12)
where
istheinputcomplexbasebandsignal.
istheoutputcomplexbasebandsignal.
,
arecomplexvaluedparameters.
isthememorydepth.
istheorderofthepolynomial.
This model considers only oddorder nonlinear terms, because the evenorder
termsareusuallyoutsideoftheoperationalbandwidthofthesignalandcanbe
19
easilyfilteredout.Thismodelconsiderspolynomialswithordersupto 2
1,
whereKisadesignparameter.
2.2.4
Comparisonsbetweennonlinearmodelswithmemory
The Volterra series is the most general model and is the most accurate one.
However, for the Volterra series to be accurate enough, the number of
parameters needed increases dramatically. This motivates the use of subsets of
the Volterra series. The most popular subsets are the Memorypolynomial,
Hammerstein,Wiener,ParallelHammersteinandParallelWienermodels.
Comparing the memorypolynomial model (2.12) with the Hammerstein model
(2.9), it can be observed that the Hammerstein model is a special case of the
Memorypolynomialmodelwhenonlytheoddpolynomialtermsareconsidered
forthenonlinearityoftheHammersteinmodeland
1,
(2.13)
ParallelWienermodelincludestheWienermodelasaspecialcase.Hammerstein
andWienermodelsarethemostspecializedwiththeleastnumberofcoefficients,
but are by no means the easiest to identify. The memory polynomial model,
however, offers a good compromise between generality and ease of parameter
estimationandimplementation[1][2].
2.3
MemoryPolynomialModel
2.3.1
Modelsimplementation
AsshowninSection2.2.4memorypolynomialprovidesagoodtradeoffbetween
accuracy and complexity. A memorypolynomial system can be expressed as
follows
Thisequationcanberewrittenasfollows
21
(2.14)
(2.15)
(2.16)
where
1,
2
(2.17)
whichmeansthat
1,
2
2
3,
2
1,
2
2
1,
5,
1
1
4
(2.18)
AblockdiagramisshowninFig.2.7andFig.2.8
Fig.2.7.Implementationofequation2.19
22
Fig.2.8Implementationof(2.17)
2.3.2
ModelsIdentification
(2.19)
and
23
(2.20)
where
istheoutputmeasureddataelements.Basebanddataisused.
isthesizeofmeasureddataset.
and
1,
3,
1,
1,
3,
3,
1,
1,
(2.21)
1,
so
2
1,
(2.22)
Letthecomplexcoefficientsberepresentedasfollows:
(2.23)
where
1,
3,
1,
(2.24)
Thenthefollowingmatrixequationholds
(2.25)
If istheestimatedparametermatrixthentohaveminimumRMSerrorbetween
the measured and simulated output can be calculated from the following
equations
Since
24
(2.26)
where
istheconjugatetransposeof ,alsoknownastheHermitiantransposeorthe
adjointmatrix.
Then(2.25)canberewrittenas
(2.27)
(2.28)
where
(2.29)
and
where
isthepseudoinversematrixof definedabove.
25
(2.30)
then
(2.31)
Thesimulatedoutputcanbecalculatedfromtheinputandestimatedparameters
asfollows:
(2.32)
(2.33)
Thentheerrorbetweenmeasuredandsimulateddatacanbedefinedas
(2.34)
is minimized.
26
LowNoiseAmplifier
3.1
Introduction
Low Noise Amplifier (LNA) is the first amplifier in the RF receiver frontend;
typically it is the first or second component after the antenna. It is designed to
increasethepowerofthereceivedsignalwhichisusuallyveryweak(couldbeas
weakas200dBm).LNAsaredesignedtoaddaslittlenoiseaspossible,suchthat
the signal to noise ratio (SNR) stays above the minimum required SNR of the
receiver.TheSNRisdefinedastheratiobetweenthewantedsignalandthenoise
andisusuallyspecifiedindBs.EveryreceiverhasaminimumSNRatitsinput,if
the SNR drops below this value, the error in the received signal will be high.
Another important performance figure for RF receivers is the noise factor F
which is defined as the
receiverwilladd.IfFismeasuredindBsinsteadoflinearscale,itwillbeknown
asthenoisefactor.ThenotationFisusuallyusedforbothnoisefigureandnoise
factor.FormoreexplanationofFseesection3.2.4.
Ideally, an LNA should introduce a linear gain and no noise to the received
signal. However, in reality electronics devices are inherently nonlinear and
rathernoisy.
Fig.3.1:Noisefigurecalculationforreceiverfrontend
27
(3.1)
From (3.1) and Fig. 3.1, it is clear that the noise figure of the overall network
dependsmainlyonthenoisefigureofthefirstcomponentwithhighgain.Thisis
typicallytheLNA.Thisimpliesthatanaccuratemodelingofthenoiseaddedby
the LNA is crucial for a good receiver model. Another important impairment
thatshouldbeincludedforthemodelingofLNAisthenonlinearityintroduced
bythesolidstatetransistors.Mismatchingeffectisalsoanimportantimpairment
which causes an increase in the bit error rate and needs to be considered for
accurateLNAmodeling[7].
In the following Sections modeling of the above mentioned impairments in
MatlabRFblocksetispresented.
3.2
LNAsModelinRFBlockset
doesnotmodelmismatchingeffects;itassumesthatthecharacteristicimpedance
is one ohm and there is no reflection due to mismatching. To overcome this
limitationthephysicallibraryisimplemented.Thislibraryusessparametersto
modelmismatchingeffectsinthefrequencydomain.
To connect blocks from physical library to mathematical library, two special
blocks are available in RF Blockset. These are the input port and output port
blocksFig.3.2.
3.2.1
InputportandOutputportBlocks
Inputportandoutputportblocksprovideconnectionbetweenthephysicaland
mathematical environments. It can be noticed from Fig. 3.2 that the connection
betweenthephysicalblocksisbidirectionaltohighlightthefactoftheincident
and reflected waves. They also provide an interface to enter a new set of
simulation parameters. Although the physical environment seems like it is
working in the frequency domain, RF Blockset actually builds an equivalent
modelinthecomplexbasebanddomain.Thismodelisbuiltdynamically,i.e.at
runtimeratherthandesigntime.ItthenconnectsthissystemtoSimulinkand
runsthesimulation.Thisprocedureisexplainedinthefollowingparagraph.
29
Fig. 3.2 Schematic illustration of the physical and mathematical environments in RF
Blockset
Theoutputportblockproducesthebasebandequivalenttimedomainresponse
of an input traveling though series of physical components. The output ports
block:
1. Partitions the RF physical components into linear and nonlinear
subsystems.
2. Extractsthecompleximpulseresponseofthelinearsubsystemsforbase
bandequivalentmodelingofRFlinearsystem.
3. ExtractsthenonlinearAM/AMandAM/PMmodelingofRFnonlinearity.
AsshowninFig.3.3,anonlinearsubsystemisimplementedbyAM/AMand
AM/PM nonlinear models built from the nonlinear parameters specified in
thephysicalblocksbetweentheinputandoutputblocks.Ifamixerblockis
included with valid phase noise data then phase noise will be added as
showninthefigure.ThenonlinearsubsystemisfurtherexplainedinSection
3.2.5.2,andthephasenoisemodelingisfurtherexplainedinchapter4.
30
Fig.3.3basebandcomplexmodelingofphysicalblockbetweeninputandoutput
ports
Tosimulateamodelthatcontainsphysicalblocks,RFBlocksetdetermines
the modeling frequencies of the physical system using parameters in the
InputPortblock.Themodelingfrequenciesarethefrequenciesatwhichthe
information is taken from the blocks to construct the baseband equivalent
model.ThentheRFBlocksetdeterminestheblockparametervaluesatthose
frequenciesandusestheinformationtocreateabasebandequivalentmodel
for timedomain simulation in Simulink. At the time of simulation output
port block uses Input port block parameters to determine the modeling
frequencies.ThesefrequenciesareinformofvectorofNelement,where N is
Finite impulse response filter length parameter in input port block. The
modeling frequencies are a function of the center frequency f , Finite
impulse response filter length N and the sample time t . This process is
illustratedinFig3.4.[6]
31
Fig.3.4FrequencybandwidthcalculationinRFBlockset[6]
ThetransferfunctioniscalculatedatPassbandforthefrequencycalculatedand
isgivenbythefollowingequation
Where
and
(3.2)
modelingfrequencies.Morespecifically
21
21
1
22
(3.3)
where
32
and
isthesourceimpedance
istheloadimpedance
aretheSparametersofatwoportnetwork
Aftercalculatingthetransferfunctionitistransferredtobasebandasfollows
(3.4)
where
BBisforBaseBand
PBisforPassband
Thenthebasebandimpulseresponseiscalculatedasfollow
(3.5)
Fig.3.5.Basebandequivalentspectrum
33
3.2.2
MismatchseffectsmodelinginRFBlockset
RFBlocksetmodelsmismatchusingtheinputsparameterdata,frequencydata
and characteristic impedance of the amplifier. Sparameter data is first
interpolated and extrapolated to the simulation frequency and then used to
calculate the incident and reflected waves. This process is explained in the
followingSection.
3.2.3
Scatteringmatrix
Thescatteringmatrix(smatrix)foracertainnetworkdescribeshowanincident
wave (or voltage signal) on one port of the network will scatter and be
distributed among other ports, Fig. 3.6. An s matrix can be constructed for
networks with any number of ports, but since two port networks are the only
onesconsideredinthisthesis,onlytwoportSmatricesareaddressed.Supposea
twoportnetworkisassumedwithincidentandreflectedvoltagewaves
respectively,where
and
isanintegerreferringtotheportnumber.Thena2by2
smatrixforthisnetworkiswrittenasfollow
11 1
12 2
21 1
22 2
(3.6)
(3.7)
Orinmatrixform
11
12
21
22
(3.8)
34
Fig.3.6IllustrationofSmatrixfora2portnetwork
ItcanbeshownthattheSmatrixisequivalenttoothernetworkingmatricessuch
as Z, Y and ABCD , in fact transformation formulas between these matrices do
exist [8]. Formulas also exist for combining one or more networks either in
cascadeorinparallel[8].TheSmatrixisalinearcharacterizationofnetworks.It
encapsulates the voltage/current characteristics of networks linearly (Ohms
law),SmatricesareveryusefulinRFcircuitsengineeringbecausetheydescribe
thewaypowerflowsinthesystemandenabletooptimizeformaximumpower
transfers.TheSparametersaredependentonfrequencybecausetheyrepresent
networks with frequency dependant components such as inductors and
capacitors.ThereforeSmatricesaremeasuredforarangeoffrequencies,andin
simulation time after determining the center frequency and bandwidth, the
correctsetofSparametersisused.
3.2.4
ModelingofNoiseinRFBlockset
There are three ways to model noise in RF physical blocks. The accuracy of
modeling depends on the availability of the noise data. These three different
waysareasfollow
1.Thesimplestwayistospecifynoisefigure,noisefactorornoisetemperature
whichareequivalentasshowninthefollowingequations
Noisefactor(F)isdefinedasfollows[7]
35
(3.9)
NoisefigureNFisthenoisefactorspecifiedindB
10log10
(3.10)
Noisetemperature canberelatedtonoisefigureas
1
(3.11)
where isthestandardroomtemperatureinKelvin(290K)
2.Amoreaccuratewayistospecifythenoisefigureasafunctionoffrequency
forthewholefrequencyrangeasnoisedependsupontheoperationalfrequency
bandwidth.
3.ThethirdwayistospecifythenoisedataastheMinimumnoisefigure(
equivalent noise resistance
),
. The RF
Blocksetwillcalculatethenoisefigurefromthenoisecorrelationmatrix
as
follows[19]
1
(3.12)
2
where
KistheBoltzmannsconstant
isthenoisetemperatureinKelvin
Thenoisefactor iscalculatedfromthecorrelationmatrixasfollows
(3.13)
36
(3.14)
where
isthenominalimpedance.
istheHerrmitianconjugationof .
Once the noise figure of each physical block is calculated for all simulation
frequenciesusingoneoftheabovemethods,theoverallnoisefigureiscalculated
from(3.1).Thenoisepowerforthewholesystemiscalculatedfromthefollowing
formula
(3.15)
whereBisthesimulationbandwidth
ThenoiseisaddedtothesystemasshowninFig.3.3.
3.2.5
ModelingofNonlinearityofLNA
Duetothenonlinearityofsolid statetransistors,LNAhasnonlineareffectsbut
sincethepowerleveloftheinputsignalistypicallysmall,memoryeffectsonthe
output signal are not severe and can be safely ignored. Therefore the static
nonlinear model included in the RF Blockset set is used for the modeling of
nonlinearityinLNA.
3.2.5.1 ModelingofnonlinearityinRFBlockset
TheamplifierblockinRFBlocksetsphysicallibrarymodelsthenonlinearityof
theLNAasfollows
37
If AMAM and AMPM data exists (for example in .AMP file) then
AMAM/AMPM nonlinearities are extracted from this data, and the IP3
andthe1dBgaincompressionpowerareeasilyextractedfromthisdata.
If no AMAM and AMPM data are included in the source file then the
nonlinearities are determined by specifying OIP3 (or IIP3) and the 1dB
gaincompressionpoint.Ifhoweverthe1dBgaincompressionpointisnot
specifiedthennonlinearitywillbecomputedasfollow[16]
1.
ConvertthespecifiedvalueintoIIP3(ifneeded).
2.
ConverttheIIP3valuefromdecibelstolinearunits.
3.
Computeascalingfactor,whichisequalto3dividedbythelinear
IIP3value.
4.
Applythescalingfactor.
5.
(3.16)
where
isthemagnitudeofthescaledsignal.
ThespecifiedthirdorderinterceptvalueisconvertedintoOIP3(if
needed)
38
2.
Thegain,OIP3,and1dBcompressiondataareconvertedtolinear,
unitlessvalues,normalizedto1voltandthereferenceimpedance
Z0.
10
(3.17)
10
10
10
(3.18)
10
(3.19)
where
GAIN is the amplifier power gain, which is derived from the
networkparameters.
OIP3istheoutputthirdorderinterceptpoint.
PCOMPistheoutputpoweratthe1dBcompressionpoint.
3.
 
 
thatdeterminestheAM/AMconversionsoftheinputsignalx.
5
1
0.2
2
0 . 10
(3.20a)
20
0
30
3
1
100.1
(3.20b)
10
0.05
(3.20c)
4.
39
3.
9.
2
3
10.
20.
5
1. 5
(3.21)
5.
TheAM/AMconversionsareappliedtotheinputsignal.
40
Mixermodeling
4.1
Introduction
According to [7] a mixer can be defined as a three port device that uses a
nonlinearortimevaryingelementtoachievefrequencyconversion.Normally
inawirelesscommunicationsystemsallsignalprocessingisdoneinbaseband
because it is easy to process low frequency signals. To be able to transmit
through the wireless channel however the signal has to be brought to a higher
frequency, which is done by modulation and up conversion in the transmitter,
thiseffecthastobeundoneinthereceiverwhichcorrespondstodemodulation
and down conversion in the receiver. Up and down conversions are done with
mixers.
Fig4.1ASchematicofadownconversionMixer.
Therearemanyapproachestomodelmixersatthesystemlevelthatdependson
which type of impairment the model addresses. Power levels of spurious
components are modeled using intermodulation tables. Nonlinearity which is
aninherentphenomenoninmixersismodeledusingIIP3and1dBcompression
point as explained in Section 3.2.3. Noise is modeled in the same way as in
Section 3.2.2 .The main impairment related to mixers that highly effect the
41
performanceoftheoverallsystemisthephasenoise.Phasenoiseisexplainedin
followingSectiontogetherwiththealgorithmusedinRFBlocksettomodelit.
4.2
ModelingofMixersinRFBlockset
TheRFBlocksetprovidesacomplexbasebandmodelformixersthatinclude
phasenoise.Theoscillatorisalsoincludedinthemixerblock,sothemixerhas
onlyinputandoutputports.Thephysicalmixermodelisviewedasatransition
tothecenterfrequencyofthesimulation.Ifthemixerissettoupconvertitwill
shift the simulation center frequency
to
where
and
are
specifiedbytheuserandifitissettodownconverteritsoutputfrequencywill
be
.Othersubsequentblockswillthenusethisnewcenterfrequencyfor
choosingtheoperationalSparametervector,Fig.4.2
42
Fig.4.2MixermodelinginRFBlockset;upconversion(up)downconversion(down)
43
4.3
ModelingofPhaseNoise
Forthemixingprocesstotakeplaceaconstantfrequencysourceisneeded.The
signal of this source is mixed with the input signal to generate up or down
converted output signal. The device which produces constant frequency signal
for mixers is called oscillator. Ideally an oscillator should produce a pure
sinusoidalsignalwiththedesiredfrequencyhoweverinpracticepuresinusoidal
isnotachievablesoanactualoscillatorsignalwillencounterbothamplitudeand
frequencyfluctuation.Thefrequencyfluctuationisaveryseriouslimitationand
cancauseseveredegradationin performanceofthesystem.Thisimpairmentis
characterizedbyPhaseNoise.
PhasenoisecanbedefinedasAshorttermrandomfluctuationinthefrequency
(Phase) of an oscillator signal [7]. Phase noise can be modeled with a simple
feedbackmodellike,LeesonsModel[15].
IntheRFBlocksethoweverarathercomplicatedmodelisusedthatdependson
1
powerlawnoisegeneration[10].Theimplementationofwhichisexplained
infollowingSectionbriefly.
4.3.1
PhasenoisemodelinginRFBlockset
The mixer block available in RF Blockset , adds phase noise in the following
order
1. An additive white noise (AWGN) ,which is correlated to the input signal is
generated
2. Thegeneratednoiseisthenfilteredwithadigitalfilter.
44
Fig.4.3phaseNoiseModelinginRFBlockset
45
46
Analysisofmemorypolynomialmodel
5.1
Introduction
In this chapter the polynomial model explained in [1] and implemented in this
thesis is tested and the results are presented and commented. The overall
transmitter and receiver were also tested and the results of which are also
presentedandcommentedinthischapter.
5.2
Modelsanalysis
dataforatypicalWIMAXamplifier .
5.2.1
AM/AMandAM/PMPlots
TheAM/AMplotsaregeneratedbyplottingoutputvoltages(orpowers)versus
input voltages (or powers). AM/PM plots on the other hand are generated by
plotting the phase differences between input and output data, versus input
voltages (or powers). Only voltage AM/AM and AM/AM are presented in this
chapter.
InFig.5.1.anAM/AMisplottedforamemorypolynomialmodelwithmemory
depth
3andpolynomialofthe5thdegree.
3,onthebackgroundaplotfor
themeasureddataisalsoplotted.
This data was provided by the microwave electronic laboratory, in the department of Microtechnology and
Nanoscience (MC2), and the communication systems group in the department of signals and systems (S2), Chalmers
University of technology. It is for a GaN power amplifier with bandwidth of 3.84 MHz. the measurement was done at
2.1 GHz center frequency and up to 5 times the bandwidth of the amplifier stated above with a maximum power of
15dBm.
47
Fig.5.1.VoltageAM/AMformeasured(blue),andmodeled(Red)datafor
3,
3.
Fromthefigureitisclearthatthemodelfitsthedataverywellatsmallsignals
(signals below 2 volts) and it gets less accurate as the operating voltage, or
power,increases.Thatcouldbeaseriouslimitationforapoweramplifiermodel
because it is typically operated in high power levels and the distortion,
introducedtothesignal,atthesepowerlevelsarethemostimportant.However,
2 volt is a typically high voltage especially for a mobile device with limited
powerresources.
Themodelwasinvestigatedalongitistwodimensions,whicharememorydepth
and polynomial order 2
versusthemodeleddatawhilethememorydepthwaskeptconstantat
thepolynomialordersfromthe3rdtotheeleventh,thatisfrom
0and
2to6.
48
Fig.5.2.VoltageAM/AMofmeasureddata(red)andmodeleddataforamemorydepth
ofzeroandpolynomialorderof3rd(yellow),toeleventh(black).
Theorderofthepolynomialdoesnotaffecttheaccuracyofthemodelverymuch
as it increased above three [1]. This can also be seen on Fig. 5.2. Therefore the
memoryorderwaskeptconstantat
3.
The model then was designed with constant polynomial order of three and
tunablememorydepthsthatcantakeanynumberbetweenzeroandsix.
It can be concluded from Fig. 5.1 and Fig. 5.2. that the implemented model fits
(simulates)themeasuredamplifierverywellifthequalityofthedataisgood.
ThememoryeffectcanbeobservedbyplottingtheAM/AMforthemodeleddata
foraconstantpolynomialorderofthree,andmemorydepthsofzeroandsix,this
is shown in Fig. 5.3. This memory effect can be better observed by applying
Gaussian noise data followed by a finite impulse response filter (FIR) and
49
plotting the AM/AM and AM/PM of the model. This is shown in 5.2.1Fig. 5.4.
form this figure the spread of the plot is more wild around the memoryless
response, this is due to the high signal fluctuation (standard deviation) which
triggersmemorymorerapidly.
Fig.5.3.VoltageAM/AMformodeleddataforpolynomialorder3andmemorydepth
zero(red)and6(blue).
(a)
(b)
Fig.5.4a)AM/PMresponseforaGaussian
noiseinput.Theredplotsareforamemory
lessmodelandtheblueonesareforamodel
withmemory.
Fig.5.4b)AM/AMresponseforaGaussian
noiseinput.Theredplotsareforamemory
lessmodelandtheblueonesareforamodel
withmemory.
50
1 maxima and
minimaandthatitapproachesinfinityforlargepositivevalues,thereforethe
modelisenforcedto saturation asshown inFig. 5.5. Thepoint of saturation
waspickedaftertheidentificationofthemodelasthepointwhentheslopeof
theAMAMcurvebecomeszeroforthefirsttime.
Fig.5.5.Modelresponseforarampinput(nomemoryeffectappears).Red:without
saturationenforced.Blue:withsaturationenforcedatmaxima.
5.2.2
Multitonetest
Atwotonesignalisappliedintheformoftwopuresinusoidalsignals,and
thefrequencyspectrumoftheoutputofthemodelisplottedinFig.5.6.from
this figure it is clear that when a memory is introduced in the model an
asymmetrybetweentheupperandlowersidebandsoftheoutputsignal.This
asymmetry was claimed to be evidence and a characteristic of systems with
memory[3].
51
Fig..5.6.Frequ
uencyspecttrumofpoly
ynomialmo
odeloutputtwhenthe inputisattwo
toneesignaland
dthemodellhas;Red:m
memorydep
pthsix.Blue:memorydepthzero.
5.2.3
Resp
ponsetoWCDMA
Asignal
AW
WCDMAsiignalwasappliedtotheampliffierandth
hefrequenccyspectrum
mof
theoutputisp
plottedforrmemoryaandmemo
orylessmo
odels,Fig.55.7.
Fig..5.7.Frequeencyspectru
umforinpu
ut(red),outtputofmem
morylessm
model(blue)and
outp
putofmodeelwithmem
mory(black
k).
52
Chaapter 5 An
nalysis of memorypo
m
olynomial model
m
Thee frequency regrowth can be seen clearrly in Fig. 5.7. the m
memory effect,
how
wever is not
n clear but
b can beetter be seeen by step
pping thro
ough different
valu
uesofmem
morydeptths,andpllotthefreq
quencyspeectrumin eachcase Fig.
5.8.
It can
c
be co
oncluded from
f
Fig. 5.8. that as the memory
m
leevel increa
ases,
inteerference with adjaacent chan
nnels will become more prrobable. That
T
com
mplieswith
htheoreticaalanalysiss.
Fig..5.8.Thep
powerspecttrumdensittyof:theinp
putsignalttothememo
orypolynomial
mod
del(red),an
ndtheoutputofthemo
odel,fordiffferentmem
morylevels..
53
54
Conclusionsandfuturework
InthisthesisabehavioralmodelsforRFfrontendcomponentsinSimulink
/RFBlocksetareinvestigatedandthealgorithmsbehindthemareexplained.
A Nonlinear model for the power amplifier that takes into account memory
effects is implemented in Simulink /RF Blockset. Simulink models are
easiertounderstandandreusebetweendifferentapplications.Designtimein
Simulink is shorter than in Matlab. However, the simulation time in
SimulinkislongerthaninMatlab,especiallyforlargeRFBlocksetmodels.
Also,SimulinkmodelsarelessflexiblethanMatlabmodels,butveryflexible
modelscanbebuiltinSimulinkusingSfunctionsandembeddedfunctions.
The accuracy of the memorypolynomial model is good, however better
accuracycanbeachievedbyusingasparsedelaymodel[1].
Toextendthereceiverfurther,amodelfortheanalogtodigitalconvertercan
be added to the receiver, this will add the effect of quantization noise and
nonlinearity. The effect of the RF filters in the receiver and transmitter can
alsobeadded.
55
56
AppendixA
DescriptionofimplementedSimulinkModels
This is a model for an RF power amplifier. It is built for wideband
applications,e.gWCDMA.Itmodelsthenonlinearityaswellasthememory
effects.Thismodelisanimplementationofthememorypolynomialmodel.
PowerAmplifierModel(P.A.withmemory)
Fig.A.1ModelblockswithSparameterpresentationinSimulink
57
Fig:A.2SchematicofthemodelafterincludingSparametercomponents.
BuildingblocksandParametersidentification:
The model is built with blocks from Simulink, RF Blockset and
communicationBlockset.Theprocessofbuildingthemodelfrommemory
polynomial equations is explained in Section 2.3.1, and process of finding
coefficientparameterisexplainedtheoreticallyinSections2.3.2,thepractical
explanationisasfollow:
Theinputandoutputdataofatypicalpoweramplifierismeasured
usinge.g.loadpullmethod
ThisdataisconvertedintoIQorcomplexbasebanddata
MATLABcodegivenAppendixBisusedtofindthecoefficientsofthe
memorypolynomialmodel.
PlugthesecoefficientsinthemodelasgiveninthePmatrixinthe
initializationfieldofthemodelsmask.
Usingmodelinsystem
MakesurethatthelibraryRFFrontendisinyourMatlabspath.
OpenSimulinkbrowser,theRFFrontendlibraryshouldbethere,
draganddropblocksfromthislibrarytoyourmodel.
Samplingtime
In Simulink blocks putting 1 in the sampling time fields means that the
block will inherit the sampling time of the preceding or subsequent blocks.
58
Runthemodel,RFBlocksetwillgivessamplingtimeerrorandwill
statethecorrectsamplingtimeofprecedingblock.
Copythissamplingtimetoyourmodelsinputport.
DialogBox
Fig.2:Dialogbox
59
Datasource
This field is used to feed network parameters (Sparameter data).The
choices aretouchstonefiles(S2P, S2D orAMP files)andsparameters
vector.Itisnottunable.
DataFile
IfDatasourcefieldissettoS2Pfile,usethisfieldtospecifythenameof
thefilethatcontainstheamplifierdata.Thefilenamemustincludethe
extension.IfthefileisnotinyourMatlabpath,specifythefullpathto
thefile.Thisfieldisnottunable.
Sparameters
If the Data source field is set to Sparameter vector, use this field to
give Sparameter in vector form;
. This field is
nottunable.
FrequencyforSparameters
IncaseiftheDatasourcefieldissettoSparametervectoroption,this
field gives the value of the frequency corresponding to the network
parametersinSparameterfield.Thisfieldisnottunable.
CarrierFrequencyHz
Thisfieldisusedtoentercarrierfrequency.Thisfieldisnottunable.
Gain(dB)
Thisisthesmallsignallineargain.Thisfieldisnottunable.
Referenceimpedance(Ohm)
It is the characteristic impedance of the physical block to which the
poweramplifierisconnected.Typicalvalueis50ohm.
Sampletime(seconds)
60
Thisfieldistosetthetimeintervalbetweenconsecutivesamplesofthe
inputsignal.Thisfieldisnottunable.
AddNoise
Ifthisfieldischeckedthennoisewillbeadded.
Noisefigure
WhenthefieldAddnoiseisenabled,thenhereyoucanputthescalar
valueofnoisefigureindecibel.Thisfieldisnottunable.
Addmemoryeffects
If checked this field will enable the addition of memory effects to the
model.Ifitisuncheckedthenthememorydepthwillbezero,i.e.memoryless
polynomial.Thisfieldisnottunable.
MemoryEffects
IfthefieldAddmemoryeffectsisenabledherethedepthofmemory
can be selected between 1 and maximum of 6. This field is not
tunable.
61
62
AppendixB
MatlabCodestocalculatecoefficients
OncetheinputandtheoutputdataforthepoweramplifierareavailableinIQ
format, the following code I is used to calculate coefficients for specified
values of memory depth m and odd polynomial order of n. Only odd
componentsofthepolynomialareconsideredsinceevencomponentscanbe
easily filtered out in the receiver. n in the code refers to the number of
polynomialtermsandtheorderofthepolynomialis2n1.CodeIIisforboth
evenandoddterms
CodeI:
functiona=analyzemodel_mp(x,y,n,m)
%[A,H]=ANALYZEMODEL_MP(X,Y,N,M)Identificationofamemorypolynomialfrom
%xandy.Nistheorderofthenonlinearity,andMthememorydepth;if
%omitted,theirdefaultvaluesare6and4,respectively.
%Normalusage:A=ANALYZEMODEL_MP(X,Y)identifiesthemodelintothematrix
%a.Eachcolumnisthecoefficientsforthenonlinearities(firstcolis
%thenonlinearitycoefffordelay0,thenfordelay1etc);eachrowis
%thefilterofeachterminthenonlinearity(row1isthefilterofx,
%row2ofx^2etc)
%notethattheidentifiedpolynomialtakesintoaccountonlyodd
%polynomialterms.
%(c)2007ThomasErikssonthomase@chalmers.se
ifnargin<4
n=6;%orderofnonlinearity
m=4;%orderofmemory
end
H=geth(x,n,m);
a=pinv(H)*y;
a=reshape(a,n,m+1);
63
functionH=geth(x,n,m)
N=length(x);
H=[];
forq=0:m
Hq=zeros(N,n);
fork=1:n
forl=(q+1):N
Hq(l,k)=abs(x(lq))^(2*k2)*x(lq);
end
end
H=[HHq];
end
CodeII:
functiona=analyze_dpd(y,z,K,Q)
%[A,H]=ANALYZE_DPD(X,Y,N,M)Identificationofamemorypolynomialfrom
%xandy.Nistheorderofthenonlinearity,andMthememorydepth.
%Normalusage:A=ANALYZE_DPD(X,Y,N,M)identifiesthemodelintothematrix
%a.Eachcolumnisthecoefficientsforthenonlinearities(firstcolis
%thenonlinearitycoefffordelay0,thenfordelay1etc);eachrowis
%thefilterofeachterminthenonlinearity(row1isthefilterofx,
%row2ofx^2etc)
%notethattheidentifiedpolynomialcontainsbothevenandoddparts.
%(c)2008EyadARABI
U=getU(y,K,Q);
a=U/z;%thisgivestheleastsquareerror%%pinv(U)*zcanalsobeused
a=reshape(a,K,Q+1);
functionU=getU(y,K,Q)
N=length(y);
U=[];
forq=0:Q
Uq=zeros(N,K);
64
MatlabcodeforMemorypolynomialmodel
The following code is an implementation of the memorypolynomial model.
ThiscodeisequivalenttotheSimulinkmodelbuiltinSection2.3.1ofthis
thesis. TheSimulinkmodelandtheMatlabcodewereverifiedagainsteach
other. In this code x is the complex input vector to the model, Y is the
complexoutputvector,andaisthecoefficientmatrixcalculatedfromabove
code.
functiony=runmodel_mp(x,a)
%(c)2007ThomasErikssonthomase@chalmers.se
n=size(a,1);%orderofnonlinearity
m=size(a,2)1;%orderofmemory
a=reshape(a,prod(size(a)),1);
H=geth(x,n,m);
y=H*a;
functionH=geth(x,n,m)
N=length(x);
H=[];
forq=0:m
Hq=zeros(N,n);
fork=1:n
forl=(q+1):N
65
Hq(l,k)=abs(x(lq))^(2*k2)*x(lq);
end
end
H=[HHq];
End
66
Reference list
Referencelist
[1]
[2]
[3]
MagnusIsaksson,DavidWisell,andDanielRnnow,AComparative
Analysis of Behavioral Models for RF Power Amplifiers, IEEE Trans
onMicrowaveTheoryAndTechniques,Vol.54,no.1,January2006
[4]
A.Ahmed,Analysis,ModelingandLinearizationofNonlinearityand
MemoryEffectsinPowerAmplifiersUsedforMicrowaveandMobile
Communications,Ph.Dthesis,UniversityofKassel,Germany,March
2005.
[5]
[6]
www.mathworks.com
[7]
DavidM.Pozar,MicrowaveandRFDesignofWirelessSystems,1st
editionJohnWiley&Sons.November2000
[8]
DavidM.Pozar,Microwaveengineering,2ndeditionJohn Wiley&
Sons.1998
[9]
[10]
[11]
Jie He , Jun Seok Yang, Yongsup Kim and Kim, A.S., SystemLevel
TimeDomainBehavioralModelingforAMobileWiMaxTransceiver,
TheProceedingsoftheIEEE,Sept2006,pp.138143.
[12]
67
[13]
[14]
[15]
D.B.Leeson,AsimplemodeloffeedbackoscillatorNoisespectrum,
Proc.IEEE,Vol.54pp.329330,1966
[16]
RFBlocksetUserguide(www.mathworks.com)
68