are used in a wide varietyof applications includingwireless communication,jamming, imaging, radar,and RF heating. Thisarticle provides an intro-duction and historical
background for the subject, and begins thetechnical discussion with material on signals,linearity, efficiency, and RF-power devices. Atthe end, there is a convenient summary of theacronyms used—this will be provided with allfour installments. Author affiliations and con-tact information are also provided at the endof each part.
1. INTRODUCTIONThe generation of significant power at RF
and microwave frequencies is required notonly in wireless communications, but also inapplications such as jamming, imaging, RFheating, and miniature DC/DC converters.Each application has its own unique require-ments for frequency, bandwidth, load, power,efficiency, linearity, and cost. RF power can begenerated by a wide variety of techniquesusing a wide variety of devices. The basictechniques for RF power amplification viaclasses A, B, C, D, E, and F are reviewed andillustrated by examples from HF through Kaband. Power amplifiers can be combined intotransmitters in a similarly wide variety ofarchitectures, including linear, Kahn, enve-
lope tracking, outphasing, and Doherty.Linearity can be improved through techniquessuch as feedback, feedforward, and predistor-tion. Also discussed are some recent develop-ments that may find use in the near future.
A power amplifier (PA) is a circuit for con-verting DC input power into a significantamount of RF/microwave output power. Inmost cases, a PA is not just a small-signalamplifier driven into saturation. There existsa great variety of different power amplifiers,and most employ techniques beyond simplelinear amplification.
A transmitter contains one or more poweramplifiers, as well as ancillary circuits such assignal generators, frequency converters, mod-ulators, signal processors, linearizers, andpower supplies. The classic architectureemploys progressively larger PAs to boost alow-level signal to the desired output power.However, a wide variety of different architec-tures in essence disassemble and thenreassemble the signal to permit amplificationwith higher efficiency and linearity.
Modern applications are highly varied.Frequencies from VLF through millimeterwave are used for communication, navigation,and broadcasting. Output powers vary from 10mW in short-range unlicensed wireless sys-tems to 1 MW in long-range broadcast trans-mitters. Almost every conceivable type of mod-ulation is being used in one system or anoth-er. PAs and transmitters also find use in sys-tems such as radar, RF heating, plasmas, laserdrivers, magnetic-resonance imaging, andminiature DC/DC converters.
With this issue, we begin afour-part series of articles
that offer a comprehensiveoverview of power amplifiertechnologies. Part 1 coversthe key topics of amplifier
linearity, efficiency andavailable RF power devices
RF and Microwave PowerAmplifier and TransmitterTechnologies — Part 1
By Frederick H. Raab, Peter Asbeck, Steve Cripps, Peter B. Kenington, Zoya B. Popovic, Nick Pothecary, John F. Sevic and Nathan O. Sokal
No single technique for poweramplification nor any single trans-mitter architecture is best for allapplications. Many of the basic tech-niques that are now coming into usewere devised decades ago, but haveonly recently been made practicalbecause of advances in RF-powerdevices and supporting circuitry suchas digital signal processing (DSP).
2. HISTORICAL DEVELOPMENTThe development of RF power
amplifiers and transmitters can bedivided into four eras:
Spark, Arc, and AlternatorIn the early days of wireless com-
munication (from 1895 to the mid1920s), RF power was generated byspark, arc, and alternator techniques.The original RF-power device, thespark gap, charges a capacitor to ahigh voltage, usually from the ACmains. A discharge (spark) throughthe gap then rings the capacitor, tun-ing inductor, and antenna, causingradiation of a damped sinusoid.Spark-gap transmitters were rela-tively inexpensive and capable ofgenerating 500 W to 5 kW from LF toMF [1].
The arc transmitter, largelyattributed to Poulsen, was a contem-porary of the spark transmitter. Thearc exhibits a negative-resistancecharacteristic which allows it to oper-ate as a CW oscillator (with somefuzziness). The arc is actually extin-guished and reignited once per RFcycle, aided by a magnetic field andhydrogen ions from alcohol drippedinto the arc chamber. Arc transmit-ters were capable of generating asmuch as 1 MW at LF [2].
The alternator is basically an ACgenerator with a large number ofpoles. Early RF alternators by Teslaand Fessenden were capable of oper-ation at LF, and a technique devel-oped by Alexanderson extended theoperation to LF [3]. The frequencywas controlled by adjusting the rota-tion speed and up to 200 kW could be
generated by a single alternator. Onesuch transmitter (SAQ) remainsoperable at Grimeton, Sweden.
Vacuum TubesWith the advent of the DeForest
audion in 1907, the thermoionic vac-uum tube offered a means of elec-tronically generating and controllingRF signals. Tubes such as the RCAUV-204 (1920) allowed the transmis-sion of pure CW signals and facilitat-ed the transition to higher frequen-cies of operation.
Younger readers may find it con-venient to think of a vacuum tube asa glass-encapsulated high-voltageFET with heater. Many of the con-cepts for modern electronics, includ-ing class-A, -B, and -C power ampli-fiers, originated early in the vacuum-tube era. PAs of this era were charac-terized by operation from high volt-ages into high-impedance loads andby tuned output networks. The basiccircuits remained relatively un-changed throughout most of the era.
Vacuum tube transmitters weredominant from the late 1920sthrough the mid 1970s. They remainin use today in some high powerapplications, where they offer a rela-tively inexpensive and rugged meansof generating 10 kW or more of RFpower.
Discrete TransistorsDiscrete solid state RF-power
devices began to appear at the end ofthe 1960s with the introduction of sil-icon bipolar transistors such as the2N6093 (75 W HF SSB) by RCA.Power MOSFETs for HF and VHFappeared in 1974 with the VMP-4 bySiliconix. GaAs MESFETs introducedin the late 1970s offered solid statepower at the lower microwave fre-quencies.
The introduction of solid-stateRF-power devices brought the use oflower voltages, higher currents, andrelatively low load resistances.Ferrite-loaded transmission linetransformers enabled HF and VHF
PAs to operate over two decades ofbandwidth without tuning. Becausesolid-state devices are temperature-sensitive, bias stabilization circuitswere developed for linear PAs. It alsobecame possible to implement a vari-ety of feedback and control tech-niques through the variety of op-amps and ICs.
Solid-state RF-power deviceswere offered in packaged or chipform. A single package might includea number of small devices. Power out-puts as high as 600 W were availablefrom a single packaged push-pulldevice (MRF157). The designer basi-cally selected the packaged devicethat best fit the requirements. Howthe transistors were made wasregarded as a bit of sorcery thatoccurred in the semiconductor housesand was not a great concern to theordinary circuit designer.
Custom/Integrated TransistorsThe late 1980s and 1990s saw a
proliferation variety of new solid-state devices including HEMT,pHEMT, HFET, and HBT, using avariety of new materials such as InP,SiC, and GaN, and offering amplifica-tion at frequencies to 100 GHz ormore. Many such devices can be oper-ated only from relatively low voltages.However, many current applicationsneed only relatively low power. Thecombination of digital signal process-ing and microprocessor control allowswidespread use of complicated feed-back and predistortion techniques toimprove efficiency and linearity.
Many of the newer RF-powerdevices are available only on a made-to-order basis. Basically, the designerselects a semiconductor process andthen specifies the size (e.g., gateperiphery). This facilitates tailoringthe device to a specific power level, aswell as incorporating it into an RFICor MMIC.
3. LINEARITYThe need for linearity is one of the
principal drivers in the design of
26 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
modern power amplifiers. Linearamplification is required when thesignal contains both amplitude andphase modulation. It can be accom-plished either by a chain of linearPAs or a combination of nonlinearPAs. Nonlinearities distort the signalbeing amplified, resulting in splatterinto adjacent channels and errors indetection.
Signals such as CW, FM, classicalFSK, and GMSK (used in GSM) haveconstant envelopes (amplitudes) andtherefore do not require linear ampli-fication. Full-carrier amplitude mod-ulation is best produced by high levelamplitude modulation of the final RFPA. Classic signals that require lin-ear amplification include single side-band (SSB) and vestigal-sideband(NTSC) television. Modern signalsthat require linear amplificationinclude shaped-pulse data modula-tion and multiple carriers.
Shaped Data Pulses Classic FSK and PSK use abrupt
frequency or phase transitions, orequivalently rectangular data pulses.The resultant RF signals have con-stant amplitude and can therefore beamplified by nonlinear PAs with goodefficiency. However, the resultantsinc-function spectrum spreads sig-nal energy over a fairly wide band-width. This was satisfactory for rela-
tively low data rates and a relativelyuncrowded spectrum.
Modern digital signals such asQPSK or QAM are typically generat-ed by modulating both I and Q sub-carriers. The requirements for bothhigh data rates and efficient utiliza-tion of the increasingly crowded spec-trum necessitates the use of shapeddata pulses. The most widely usedmethod is based upon a raised-cosinechannel spectrum, which has zerointersymbol interference duringdetection and can be made arbitrari-ly close to rectangular [4]. A raised-cosine channel spectrum is achievedby using a square-root raised-cosine(SRRC) filter in both the transmitterand receiver. The resultant SRRCdata pulses (Figure 1) are shapedsomewhat like sinc functions whichare truncated after several cycles. Atany given time, several different datapulses contribute to the I and Q mod-ulation waveforms. The resultantmodulated carrier (Figure 2) hassimultaneous amplitude and phasemodulation with a peak-to-averageratio of 3 to 6 dB.
Multiple Carriers and OFDMApplications such as cellular base
stations, satellite repeaters, andmulti-beam “active-phased-array”transmitters require the simultane-ous amplification of multiple signals.
Depending on the application, thesignals can have different ampli-tudes, different modulations, andirregular frequency spacing.
In a number of applicationsincluding HF modems, digital audiobroadcasting, and high-definitiontelevision, it is more convenient touse a large number of carriers withlow data rates than a single carrierwith a high data rate. The motiva-tions include simplification of themodulation/demodulation hardware,equalization, and dealing with multi-path propagation. Such OrthogonalFrequency Division Multiplex(OFDM) techniques [5] employ carri-ers with the same amplitude andmodulation, separated in frequencyso that modulation products from onecarrier are zero at the frequencies ofthe other carriers.
Regardless of the characteristicsof the individual carriers, the resul-tant composite signal (Figure 2) hassimultaneous amplitude and phasemodulation. The peak-to-averageratio is typically in the range of 8 to13 dB.
NonlinearityNonlinearities cause imperfect
reproduction of the amplified signal,resulting in distortion and splatter.Amplitude nonlinearity causes theinstantaneous output amplitude or
Figure 1 · SRRC data pulses. Figure 2 · RF waveforms for SRRC and multicarrier signals.
28 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
envelope to differ in shape from thecorresponding input. Such nonlinear-ities are the variable gain or satura-tion in a transistor or amplifier.Amplitude-to-phase conversion is aphase shift associated with the signalamplitude and causes the introduc-tion of unwanted phase modulationinto the output signal. Amplitude-to-phase conversion is often associatedwith voltage-dependent capacitancesin the transistors. While imperfectfrequency response also distorts asignal, it is a linear process andtherefore does not generate out-of-band signals.
Amplitude nonlinearity andamplitude-to-phase conversion aredescribed by transfer functions that
act upon the instantaneous signalvoltage or envelope. However, memo-ry effects can also occur in high-power PAs because of thermal effectsand charge storage. Thermal effectsare somewhat more noticeable in III-V semiconductors because of lowerthermal conductivity, while charge-storage effects are more prevalent inoverdriven BJT PAs.
Measurement of LinearityLinearity is characterized, mea-
sured, and specified by various tech-niques depending upon the specificsignal and application. The linearityof RF PAs is typically characterizedby C/I, NPR, ACPR, and EVM(defined below).
The traditionalmeasure of lineari-ty is the carrier-to-intermodulation(C/I) ratio. The PAis driven with twoor more carriers(tones) of equala m p l i t u d e s .N o n l i n e a r i t i e scause the produc-tion of intermodu-lation products atfrequencies corre-sponding to sumsand differences ofmultiples of thecarrier frequencies
[6]. The amplitude of the third-orderor maximum intermodulation distor-tion (IMD) product is compared tothat of the carriers to obtain the C/I.A typical linear PA has a C/I of 30 dBor better.
Noise-Power Ratio (NPR) is a tra-ditional method of measuring the lin-earity of PAs for broadband andnoise-like signals. The PA is drivenwith Gaussian noise with a notch inone segment of its spectrum.Nonlinearities cause power to appearin the notch. NPR is the ratio of thenotch power to the total signal power.
Adjacent Channel Power Ratio(ACPR) characterizes how nonlinear-ity affects adjacent channels and iswidely used with modern shaped-pulse digital signals such as NADCand CDMA. Basically, ACPR is theratio of the power in a specified bandoutside the signal bandwidth to therms power in the signal (Figure 3).In some cases, the actual power spec-trum S(f) is weighted by the frequen-cy response H(f) of the pulse-shapingfilter; i.e. (eq. 1)
W-CDMA (3G-PP) 5.00 MHz 10.0 MHz 4.68 MHz RRC 25%/N/A [18] –33 dB –43 dB α=0.22
Table 1 · ACPR and EVM requirements of various systems.
May 2003 29
where fc is the center frequency, B isthe bandwidth, fo is the offset, and fLand fU are the band edges. Theweighting, frequency offsets, andrequired ACPRs vary with applica-tion as shown in Table 1. ACPR canbe specified for either upper or lowersideband. In many cases, two differ-ent ACPRs for two different frequen-cy offsets are specified. ACPR2, basedupon the outer band, is sometimescalled “Alternate Channel PowerRatio.”
Error Vector Magnitude (EVM) isa convenient measure of how nonlin-earity interferes with the detectionprocess. EVM is defined (Figure 4) asthe distance between the desired andactual signal vectors, normalized to afraction of the signal amplitude.Often, both peak and rms errors arespecified (Table 1).
4. EFFICIENCYEfficiency, like linearity, is a criti-
cal factor in PA design. Three defini-tions of efficiency are commonly used.Drain efficiency is defined as theratio of RF output power to DC inputpower:
η = Pout/Pin (2)
Power-added efficiency (PAE)incorporates the RF drive power bysubtracting it from the output power;
i.e. (Pout – PDR)/Pin. PAE gives a rea-sonable indication of PA performancewhen gain is high; however, it canbecome negative for low gains. Anoverall efficiency such as Pout/(Pin +PDR) is useable in all situations. Thisdefinition can be varied to includedriver DC input power, the powerconsumed by supporting circuits, andanything else of interest.
Average EfficiencyThe instantaneous efficiency is
the efficiency at one specific outputlevel. For most PAs, the instanta-neous efficiency is highest at thepeak output power (PEP) anddecreases as output decreases.Signals with time-varying ampli-tudes (amplitude modulation) there-fore produce time-varying efficien-cies. A useful measure of performanceis then the average efficiency, whichis defined [7] as the ratio of the aver-age output power to the average DC-input power:
ηAVG = PoutAVG/PinAVG (3)
This concept can be used with anyof the three definitions of efficiency.
The probability-density function(PDF) of the envelope gives the rela-tive amount of time an envelopespends at various amplitudes (Figure5). Also used is the cumulative distri-
bution function (CDF), which givesthe probability that the envelopedoes not exceed a specified ampli-tude. CW, FM, and GSM signals haveconstant envelopes and are thereforealways at peak output. SRRC datamodulation produces PDFs that areconcentrated primarily in the upperhalf of the voltage range and havepeak-to-average ratios on the order of3 to 6 dB. Multiple carriers [8] pro-duce random-phasor sums much likerandom noise and therefore haveRayleigh-distributed envelopes; i.e.,
p(E) = 2E ξ exp(–V2 ξ) (4)
Peak-to-average ratio ξ is typicallybetween 6 and 13 dB.
The average input and outputpowers are found by integrating theproduct of their variation with ampli-tude and the PDF of the envelope.Two cases are of special interest.When the DC input current is con-stant (class-A bias), the DC inputpower is also constant. The averageefficiency is then ηPEP/ξ. If the DCinput current (hence power) is pro-portional to the envelope (as in class-B), the average efficiency is (4/π)1/2
ηPEP, for a Rayleigh-distributed sig-nal. Thus for a multicarrier signalwith a 10 dB peak-to-average ratio,ideal class-A and B PAs with PEPefficiencies of 50 and 78.5 percent,
respectively, have average efficienciesof only 5 and 28 percent, respectively.
Back-OffThe need to conserve battery
power and to avoid interference toother users operating on the samefrequency necessitates the transmis-sion of signals whose peak ampli-tudes well below the peak outputpower of the transmitter. Since peakpower is needed only in the worst-case links, the “back-off” is typicallyin the range of 10 to 20 dB.
For a single-carrier mobile trans-mitter, back-off rather than envelopePDF is dominant in determining theaverage power consumption andaverage efficiency. The PDF of thetransmitting power (Figure 4)depends not only upon the distance,but also upon factors such as attenu-ation by buildings, multipath, andorientation of the mobile antenna [8],[9], [10]. To facilitate prediction of thepower consumption, the envelope andback-off PDFs can be combined [11].
5. RF POWER TRANSISTORSA wide variety of active devices is
currently available for use in RF-power amplifiers, and RF-powertransistors are available in packaged,die, and grown-to-order forms.Packaged devices are used at fre-quencies up to X band, and are domi-nant for high power and at VHF andlower frequencies. A given packagecan contain one or more die connect-ed in parallel and can also includeinternal matching for a specific fre-quency of operation. Dice (chips) canbe wire-bonded directly into a circuitto minimize the effects of the packageand are used up to 20 GHz. InMMICs, the RF-power device isgrown to order, allowing its size andother characteristics to be optimizedfor the particular application. Thisform of construction is essential forupper-microwave and millimeter-wave frequencies to minimize theeffects of strays and interconnects.Virtually all RF power transistors
are npn or n-channel types becausethe greater mobility of electrons (vs.holes) results in better operation athigher frequencies.
Bipolar Junction Transistor (BJT)The Si BJT is the original solid-
state RF power device, originating inthe 1960s. Since the BJT is a verticaldevice, obtaining a high collector-breakdown voltage is relatively sim-ple and the power density is veryhigh. Si BJTs typically operate from28 V supplies and remain in use atfrequencies up to 5 GHz, especially inhigh-power (1 kW) pulsed applica-tions such as radar. While Si RFpower devices have higher gain athigh frequencies, their fundamentalproperties are basically those of ordi-nary bipolar transistors. The positivetemperature coefficient of BJTs tendsto allow current hogging, hot-spot-ting, and thermal runaway, necessi-tating carefully regulated base bias.Since RF power BJTs are generallycomposed of multiple, small BJTs(emitter sites), emitter ballasting(resistance) is generally employed toforce even division of the currentwithin a given package.
MOSFETs are constructed withinsulated gates. Topologies with bothvertical and later current flow areused in RF applications, and most areproduced by a double-diffusion pro-cess. Because the insulated gate con-ducts no DC current, MOSFETs arevery easily biased.
The negative temperature coeffi-cient of a MOSFET causes its draincurrent to decrease with tempera-ture. This prevents thermal runawayand allows multiple MOSFETs to beconnected in parallel without ballast-ing. The absence of base-charge stor-age time allows fast switching andalso eliminates a mechanism for sub-harmonic oscillation. An overdriven(saturated) MOSFET can conductdrain current in either direction,
which is very useful in switching-mode operation with reactive loads.
Vertical RF power MOSFETs areuseable through VHF and UHF.Gemini-packaged devices can deliverup to 1 kW at HF and 100s of wattsat VHF. VMOS devices typically oper-ate from 12, 28, or 50-V supplies,although some devices are capable ofoperation from 100 V or more.
Laterally Diffused MOS (LDMOS)LDMOS is especially useful at
UHF and lower microwave frequen-cies because direct grounding of itssource eliminates bond-wire induc-tance that produces negative feed-back and reduces gain at high fre-quencies. This also eliminates theneed for the BeO insulating layercommonly used in other RF-powerMOSFETs.
LDMOS devices typically operatefrom 28-V supplies and are currentlyavailable with power outputs of 120W at 2 GHz. They are relatively lowin cost compared to other devices forthis frequency range and are current-ly the device of choice for use in high-power transmitters at 900 MHz and 2GHz.
Junction FET (JFET)JFETs for power applications are
often called Static InductionTransistors (SITs). Impressive powerand efficiency have been obtainedfrom RF JFETs based upon Si, SiGe,and SiC at frequencies through UHF.However, the JFET has never becomeas popular as other RF-power FETs.
GaAs MEtal Semiconductor FET(GaAs MESFET)
GaAs MESFETs are JFETs basedupon GaAs and a Schottky gate junc-tion. They have higher mobility thando Si devices and are therefore capa-ble of operating efficiently at higherfrequencies. GaAs MESFETs arewidely used for the production ofmicrowave power, with capabilities ofup 200 W at 2 GHz and 40 W at 20GHz in packaged devices. These
32 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
devices have relatively low break-down voltages compared toMOSFETs or JFETs and are typical-ly operated from supply voltages(drain biases) of 5 to 10 V. MostMESFETs are depletion-modedevices and require a negative gatebias, although some enhance-modedevices that operate with a positivebias have been developed. Linearityis often poor due to input capacitancevariation with voltage; the outputcapacitance is also often stronglybias- and frequency-dependent.
Heterojunction FET (HFET) / High-Electron-Mobility Transistor (HEMT)
HFETs and HEMTs improve uponthe MESFET geometry by separatingthe Schottky and channel functions.Added to the basic MESFET struc-ture is a heterojunction consisting ofan n-doped AlGaAs Schottky layer,an undoped AlGaAs spacer, and anundoped GaAs channel. The disconti-nuity in the band gaps of AlGaAs andGaAs causes a thin layer of electrons(“two-dimensional electron gas or 2-DEG”) to form below the gate at theinterface of the AlGaAs and GaAslayers. Separation of the donors fromthe mobile electrons reduces colli-sions in the channel, improving themobility, and hence high-frequencyresponse, by a factor of about two.
AlGaAs has crystal-lattice proper-ties similar to those of GaAs, and thismakes it possible to produce a poten-tial difference without lattice stress.The GaAs buffer contributes to a rel-atively high breakdown voltage.Their fabrication employs advancedepitaxial technologies (MolecularBeam Epitaxy or Metal OrganicChemical Vapor Deposition) whichtends to increase their cost.
The GaAs HEMT is known in theliterature by a wide variety of differ-ent names, including MODFET(Modulation-Doped FET), TEGFET(Two-dimensional Electron-GasFET), and SDFET (Selectively DopedFET). It is also commonly called anHFET (Heterostructure FET),
although technically an “HFET” hasa doped channel that provides thecarriers (instead of the heterojunc-tion). The acronyms “HFET and“HJFET” (HeteroJunction FET)appear to be used interchangeably.
GaAs HEMTs/HFETs with fT ashigh as 158 GHz are reported. PAsbased upon these HEMTs exhibit 15-W outputs at 12 GHz with a power-added efficiency (PAE) of 50 percent.Outputs of 100 W are available at Sband from packaged devices.
Pseudomorphic HEMTThe pseudomorphic HEMT
(pHEMT) further improves upon thebasic HEMT by employing anInGaAs channel. The increasedmobility of In with respect to GaAsincreases the bandgap discontinuityand therefore the number of carriersin the two-dimensional electron gas.The lattice mismatch between theGaInAs channel and GaAs substrateis also increased, however, and thislimits the In content to about 22 per-cent.
The efficiency of PAs usingpHEMTs does not begin to drop untilabout 45 GHz and pHEMTs are use-able to frequencies as high as 80GHz. Power outputs vary from 40 Wat L band to 100 mW at V band.While pHEMTs are normally grownto order, a packaged device pHEMThas recently become available.
InP HEMTThe InP HEMT places an
AlInAs/GaInAs heterojunction on anInP substrate. The lattices are moreclosely matched, which allows an Incontent of up to about 53 percent.This results in increased mobility,which in turn results in increasedelectron velocity, increased conduc-tion-band discontinuity, increasedtwo-dimensional electron gas, andhigher transconductance. The ther-mal resistance is 40 percent lowerthan that of a comparable devicebuilt on a GaAs substrate.
The InP HEMT has higher gain
and efficiency than the GaAspHEMT, with the PA efficiency begin-ning to drop at 60 GHz. However, ithas a lower breakdown voltage (typi-cally 7 V) and must therefore be oper-ated from a relatively low drain-volt-age supply (e.g., 2 V). This results inlower output per device and possiblyloss in the combiners required toachieve a specified output power.Nonetheless, the InP HEMT general-ly has a factor-of-two efficiencyadvantage over the pHEMT andGaAs HEMT.
InP HEMTs have been fabricatedwith fmax as high as 600 GHz (0.1µm gate length), and amplificationhas been demonstrated at frequen-cies as high as 190 GHz. The efficien-cy does not begin to drop until about60 GHz. Power levels range from 100to 500 mW per chip.
Metamorphic HEMT (mHEMT)The mHEMT allows channels
with high-In content to be built onGaAs substrates. The higher electronmobility and higher peak saturationvelocity result in higher gain than ispossible in a pHEMT. mHEMTs aregenerally limited to low-power appli-cations by their relatively low break-down voltage (<3 V). However, anmHEMT capable of 6-V operationand a power output of 0.5 W has beenrecently reported.
Heterojunction Bipolar Transistor(HBT)
HBTs are typically based uponthe compound-semiconductor materi-al AlGaAs/GaAs. The AlGaAs emitteris made as narrow as possible to min-imize base resistance. The base is athin layer of p GaAs. The barrier iscreated by heterojunction (AlGaAs/GaAs) rather than the doping. Thebase can therefore be doped heavilyto minimize its resistance. Base sheetresistance is typically two orders ofmagnitude lower than that of anordinary BJT, and the frequency ofoperation is accordingly higher. Thecurrent flow is (in contrast to a MES-
34 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
FET) vertical so that surface imper-fections have less effect upon perfor-mance. The use of a semi-insulatingsubstrate and the higher electronmobility result in reduced parasitics.The DC curves are somewhat similarto those of a conventional BJT, butoften contain a saturation resistanceas well as saturation voltage.Currently available AlGaAs/GaAsHBTs are capable of producing sever-al watts and are widely used in wire-less handsets, GaAs HBTs are alsowidely used in MMIC circuits at fre-quencies up to X band and can oper-ate in PAs at frequencies as high as20 GHz.
SiGe HBTThe use of SiGe rather than Si in
the base of the HBT both increasesthe maximum operating frequencyand decreases the base resistance.However, they are generally less effi-cient than GaAs HBTs and can havelower breakdown voltages. Oneexperimental SiGe HBT is capable ofdelivering over 200 W at L band.
InP HBTThe use of InP in an HBT further
boosts mobility and therefore thehigh frequency response. In addition,InP HBTs have lower turn-on andknee voltages, resulting in highergain and efficiency. InP HBTs for RF-power applications incorporate twoheterojunctions (AlInAs/GaInAs andGaInAs/InP). The InP in the collectorincreases the breakdown voltage,allowing higher output power. Todate, outputs of about 0.5 W at 20GHz have been demonstrated, but itis anticipated that operation to 50 or60 GHz will be possible.
SiC MESFETThe wide band gap of SiC results
in both high mobility and high break-down voltage. An SiC MESFET cantherefore have a frequency responsecomparable to that of a GaAs MES-FET, but breakdown voltages doublethat of Si LDMOS. This results in
power densities of 10 W/mm, which isten times that of a GaAs MESFET.The high thermal conductivity of theSiC substrate is particularly usefulin high-power applications. The high-er operating voltage and associatedhigher load impedance greatly sim-plify output networks and powercombining. SiC MESFETs typicallyoperate from a 48-V supply. Deviceswith outputs of 10 W are currentlyavailable, and outputs of 60 W ormore have been demonstrated exper-imentally. The cost of SiC devices isat presently about ten times that ofSi LDMOS.
GaN HEMTGaN offers the same high break-
down voltage of SiC, but even highermobility. Its thermal conductivity is,however, lower, hence GaN devicesmust be built substrate such as SiCor diamond. While the GaN HEMToffers the promise of a high-power,high-voltage device operating at fre-quencies of 10 GHz or more, it is stillin an experimental state. Power out-puts of 8 W at 10 GHz with 30 per-cent efficiency have been demonstrat-ed.
Monolithic Microwave IntegratedCircuit (MMIC)
MMICs integrate RF powerdevices and matching/decoupling ele-ments such as on-chip inductors,capacitors, resistors, and transmis-sion lines. The proximity of these ele-ments to the RF-power devices isessential for input, output, and inter-stage matching at microwave andmillimeter-wave frequencies.
References1. W. J. Bryon, “Arcs and sparks,
Part 1,” Communications Quarterly,vol. 4, no. 2, pp. 27-43, Spring 1994.
2. W. J. Bryon, “The arc method ofproducing continuous waves,”Communications Quarterly, vol. 8, no.3, pp. 47-65, Summer 1998.
3. K. M. MacIlvain and W. H.Freedman, Radio Library, Vol. III:
Radio Transmitters and CarrierCurrents, Scranton PA: InternationalTextbook Company, 1928.
4. J. B. Groe and L. E. Larson,CDMA Mobile Radio Design,Norwood, MA: Artech House, 2000.
5. R. van Nee and R. Prasad,OFDM for Wireless MultimediaCommunications, Norwood, MA:Artech House, 2000.
6. H. L. Krauss, C. W. Bostian, andF. H. Raab, Solid State RadioEngineering, New York: Wiley, 1980.
7. F. H. Raab, “Average efficiencyof power amplifiers,” Proc. RFTechnology Expo '86, Anaheim, CA,pp. 474-486, Jan. 30-Feb. 1, 1986.
8. N. Pothecary, “Feedforward lin-ear power amplifiers,” in WorkshopNotes WFB, Int’l. Microwave Symp.,Boston, MA, June 16, 2001.
9. J. F. Sevic, “Statistical charac-terization of RF power amplifier effi-ciency for wireless communicationsystems,” Proc. Wireless Commun.Conf., Boulder, CO, pp. 1-4, Aug. 1997.
10. G. Hanington, P.-F. Chen, P. M.Asbeck, and L. E. Larson, “High-effi-ciency power amplifier using dynam-ic power-supply voltage for CDMAapplications,” IEEE Trans.Microwave Theory Tech., vol. 47, no. 8,pp. 1471-1476, Aug. 1999.
11. I. Kipnis, “Refining CDMAmobile-phone power control,”Microwaves & RF, vol. 39, no. 6, pp.71-76, June 2000.
12. J. Staudinger, “Applyingswitched gain stage concepts toimprove efficiency and linearity formobile CDMA power amplification,”Microwave Journal, vol. 43, no. 9, pp.152-162, Sept. 2000.
Table 1 References13. “Mobile station – base station
interoperability standard for dual-mode cellular system,” ANSI-136Standard, Telecommun. IndustriesAssoc., 2000.
14. “Digital cellular communica-tion systems,” RCR STD-27, Researchand Development Center for RadioSystems (RCR), April 1991.
36 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
15. “Digital cellular telecommunications system(phase 2+), radio transmission and reception,” GSM 5.05Standard, v. 8.4.1, European Telecommun. StandardsInst., 1999.
17. “Mobile station – base station interoperabilitystandard for dual-mode wideband spread-spectrum cellu-lar system,” TIA/EIA IS-95 Interim Standard,Telecommun. Industries Assoc., July 1993.
18. “UE radio transmission and reception (FDD),” TS25.101, v. 3.4.1, Third Generation Partnership Project,Technical Specification Group, 1999.
Series Notes1. The remaining three parts of this series will appear
in successive issues of High Frequency Electronics (July,September and November 2003 issues).
2. To maintain continuity, all figures, tables, equationsand references will be numbered sequentially throughoutthe entire series.
3. Like all articles in High Frequency Electronics, thisseries will be archived and available for downloading (forpersonal use by individuals only) online at the magazinewebsite: www.highfrequencyelectronics.com
Author InformationThe authors of this series of articles are: Frederick H.
Raab (lead author), Green Mountain Radio Research, e-mail: [email protected]; Peter Asbeck, University ofCalifornia at San Diego; Steve Cripps, HywaveAssociates; Peter B. Kenington, Andrew Corporation;Zoya B. Popovic, University of Colorado; Nick Pothecary,Consultant; John F. Sevic, California EasternLaboratories; and Nathan O. Sokal, Design Automation.Readers desiring more information should contact thelead author.
AC Alternating CurrentACPR Adjacent-Channel Power RatioBJT Bipolar-Junction TransistorC/I Carrier-to-IntermodulationCDF Cumulative Distribution FunctionCDMA Code-Division Multiple AccessCW Continuous WaveDC Direct CurrentDSP Digital Signal ProcessingEVM Error-Vector MagnitudeFET Field-Effect TransistorFSK Frequency-Shift KeyingGMSK Gaussian Minimum Shift KeyingGSM Global System for Mobile communicationHBT Heterojunction bipolar transistorHEMT High Electron-Mobility TransistorHFET Heterojunction FET (also HJFET)IC Integrated CircuitJFET Junction Field-Effect TransistorLDMOS Laterally Diffused MOS (FET)
MESFET MEtal Semiconductor FETmHEMT Metamorphic HEMTMMIC Microwave Monolithic Integrated CircuitMOSFET Metal-Oxide-Silicon Field-Effect TransistorNADC North American Digital CellularNPR Noise-Power RatioNTSC National Television Standards CommitteeOFDM Orthogonal Frequency-Division MultiplexPA Power AmplifierPAE Power-Added EfficiencyPDF Probability-Density FunctionPEP Peak-Envelope PowerpHEMT Pseudomorphic HEMTPSK Phase-Shift KeyingQAM Quadrature Amplitude ModulationQPSK Quadrature Phase Shift KeyingRF Radio FrequencySRRC Square-Root Raised CosineSSB Single SideBand
Acronyms Used in Part 1
22 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
RF and Microwave PowerAmplifier and TransmitterTechnologies — Part 2
By Frederick H. Raab, Peter Asbeck, Steve Cripps, Peter B. Kenington,Zoya B. Popovich, Nick Pothecary, John F. Sevic and Nathan O. Sokal
Part 1 of this seriesintroduced basicconcepts, discussed
the characteristics of sig-nals to be amplified, andgave background infor-mation on RF powerdevices. Part 2 reviewsthe basic techniques, rat-ings, and implementation
methods for power amplifiers operating at HFthrough microwave frequencies.
6a. BASIC TECHNIQUES FOR RF POWER AMPLIFICATION
RF power amplifiers are commonly desig-nated as classes A, B, C, D, E, and F [19]. Allbut class A employ various nonlinear, switch-ing, and wave-shaping techniques. Classes ofoperation differ not in only the method ofoperation and efficiency, but also in theirpower-output capability. The power-outputcapability (“transistor utilization factor”) isdefined as output power per transistor nor-malized for peak drain voltage and current of1 V and 1 A, respectively. The basic topologies(Figures 7, 8 and 9) are single-ended, trans-former-coupled, and complementary. Thedrain voltage and current waveforms of select-ed ideal PAs are shown in Figure 10.
Class AIn class A, the quiescent current is large
enough that the transistor remains at alltimes in the active region and acts as a cur-rent source, controlled by the drive.
Consequently, the drain voltage and currentwaveforms are (ideally) both sinusoidal. Thepower output of an ideal class-A PA is
Po = Vom2 / 2R (5)
where output voltage Vom on load R cannotexceed supply voltage VDD. The DC-powerinput is constant and the efficiency of an idealPA is 50 percent at PEP. Consequently, theinstantaneous efficiency is proportional to thepower output and the average efficiency isinversely proportional to the peak-to-averageratio (e.g., 5 percent for x = 10 dB). The uti-lization factor is 1/8.
For amplification of amplitude-modulatedsignals, the quiescent current can be varied inproportion to the instantaneous signal enve-lope. While the efficiency at PEP isunchanged, the efficiency for lower ampli-
Our multi-part series onpower amplifier tech-
nologies and applicationscontinues with a review of
tudes is considerably improved. In anFET PA, the implementationrequires little more than variation ofthe gate-bias voltage.
The amplification process in classA is inherently linear, hence increas-ing the quiescent current or decreas-ing the signal level monotonicallydecreases IMD and harmonic levels.Since both positive and negativeexcursions of the drive affect thedrain current, it has the highest gainof any PA. The absence of harmonicsin the amplification process allowsclass A to be used at frequencies closeto the maximum capability (fmax) ofthe transistor. However, the efficiencyis low. Class-A PAs are therefore typ-ically used in applications requiringlow power, high linearity, high gain,broadband operation, or high-fre-quency operation.
The efficiency of real class-A PAsis degraded by the on-state resistance
or saturation voltage of the transis-tor. It is also degraded by the pres-ence of load reactance, which inessence requires the PA to generatemore output voltage or current todeliver the same power to the load.
Class BThe gate bias in a class-B PA is
set at the threshold of conduction sothat (ideally) the quiescent drain cur-rent is zero. As a result, the transis-tor is active half of the time and thedrain current is a half sinusoid.Since the amplitude of the drain cur-rent is proportional to drive ampli-tude and the shape of the drain-cur-rent waveform is fixed, class-B pro-vides linear amplification.
The power output of a class-B PAis controlled by the drive level andvaries as given by eq. (5). The DC-input current is, however, proportion-al to the drain current which is in
turn proportional to the RF-outputcurrent. Consequently, the instanta-neous efficiency of a class-B PAvaries with the output voltage andfor an ideal PA reaches π/4 (78.5 per-cent) at PEP. For low-level signals,class B is significantly more efficientthan class A, and its average efficien-cy can be several times that of class Aat high peak-to-average ratios (e.g.,28 vs. 5 percent for ξ = 10 dB). Theutilization factor is the same 0.125 ofclass A.
In practice, the quiescent currentis on the order of 10 percent of thepeak drain current and adjusted tominimize crossover distortion causedby transistor nonlinearities at lowoutputs. Class B is generally used ina push-pull configuration so that thetwo drain-currents add together toproduce a sine-wave output. At HFand VHF, the transformer-coupledpush-pull topology (Figure 8) is gen-erally used to allow broadband oper-ation with minimum filtering. Theuse of the complementary topology(Figure 9) has generally been limitedto audio, LF, and MF applications bythe lack of suitable p-channel tran-sistors. However, this topology isattractive for IC implementation andhas recently been investigated forlow-power applications at frequen-cies to 1 GHz [20].
Class CIn the classical (true) class-C PA,
the gate is biased below threshold sothat the transistor is active for lessthan half of the RF cycle (Figure 10).Linearity is lost, but efficiency isincreased. The efficiency can beincreased arbitrarily toward 100 per-cent by decreasing the conductionangle toward zero. Unfortunately,this causes the output power (utiliza-tion factor) to decrease toward zeroand the drive power to increasetoward infinity. A typical compromiseis a conduction angle of 150° and anideal efficiency of 85 percent.
The output filter of a true class-CPA is a parallel-tuned type that
bypasses the harmonic componentsof the drain current to ground with-out generating harmonic voltages.When driven into saturation, effi-ciency is stabilized and the outputvoltage locked to supply voltage,allowing linear high-level amplitudemodulation.
Classical class C is widely used inhigh-power vacuum-tube transmit-ters. It is, however, little used insolid-state PAs because it requireslow drain resistances, making imple-mentation of parallel-tuned outputfilters difficult. With BJTs, it is alsodifficult to set up bias and drive toproduce a true class-C collector-cur-rent waveform. The use of a series-tuned output filter results in amixed-mode class-C operation that ismore like mistuned class E than trueclass C.
Class DClass-D PAs use two or more tran-
sistors as switches to generate asquare drain-voltage waveform. Aseries-tuned output filter passes onlythe fundamental-frequency compo-nent to the load, resulting in poweroutputs of (8/π2)VDD
2/R and(2/π2)VDD
2/R for the transformer-cou-pled and complementary configura-tions, respectively. Current is drawnonly through the transistor that ison, resulting in a 100-percent effi-ciency for an ideal PA. The utilizationfactor (1/2π = 0.159) is the highest ofany PA (27 percent higher than thatof class A or B). A unique aspect ofclass D (with infinitely fast switch-ing) is that efficiency is not degradedby the presence of reactance in theload.
Practical class-D PAs suffer fromlosses due to saturation, switchingspeed, and drain capacitance. Finiteswitching speed causes the transis-tors to be in their active regions whileconducting current. Drain capaci-tances must be charged and dis-charged once per RF cycle. The asso-ciated power loss is proportional toVDD
3/2 [21] and increases directly
with frequency.Class-D PAs with power outputs
of 100 W to 1 kW are readily imple-mented at HF, but are seldom usedabove lower VHF because of lossesassociated with the drain capaci-tance. Recently, however, experimen-tal class-D PAs have been tested withfrequencies of operation as high as 1GHz [22].
Class EClass E employs a single transis-
tor operated as a switch. The drain-voltage waveform is the result of thesum of the DC and RF currentscharging the drain-shunt capaci-tance. In optimum class E, the drainvoltage drops to zero and has zeroslope just as the transistor turns on.The result is an ideal efficiency of 100percent, elimination of the lossesassociated with charging the draincapacitance in class D, reduction ofswitching losses, and good toleranceof component variation.
Optimum class-E operationrequires a drain shunt susceptance0.1836/R and a drain series reac-tance 1.15R and delivers a power out-put of 0.577VDD
2/R for an ideal PA[23]. The utilization factor is 0.098.Variations in load impedance andshunt susceptance cause the PA todeviate from optimum operation [24,25], but the degradations in perfor-mance are generally no worse thanthose for class A and B.
The capability for efficient opera-tion in the presence of significantdrain capacitance makes class E use-ful in a number of applications. Oneexample is high-efficiency HF PAswith power levels to 1 kW based uponlow-cost MOSFETs intended forswitching rather than RF use [26].Another example is the switching-mode operation at frequencies ashigh as K band [27]. The class-DE PA[28] similarly uses dead-spacebetween the times when its two tran-sistors are on to allow the load net-work to charge/discharge the draincapacitances.
Class FClass F boosts both efficiency and
output by using harmonic resonatorsin the output network to shape thedrain waveforms. The voltage wave-form includes one or more odd har-monics and approximates a squarewave, while the current includes evenharmonics and approximates a halfsine wave. Alternately (“inverse classF”), the voltage can approximate ahalf sine wave and the current asquare wave. As the number of har-monics increases, the efficiency of anideal PA increases from the 50 per-cent (class A) toward unity (class D)and the utilization factor increasesfrom 1/8 (class A) toward 1/2π (classD) [29].
The required harmonics can inprinciple be produced by current-source operation of the transistor.However, in practice the transistor isdriven into saturation during part ofthe RF cycle and the harmonics areproduced by a self-regulating mecha-nism similar to that of saturatingclass C. Use of a harmonic voltagerequires creating a high impedance(3 to 10 times the load impedance) atthe drain, while use of a harmoniccurrent requires a low impedance(1/3 to 1/10 of the load impedance).While class F requires a more com-plex output filter than other PAs, theimpedances must be correct at only afew specific frequencies. Lumped-ele-ment traps are used at lower fre-quencies and transmission lines areused at microwave frequencies.Typically, a shorting stub is placed aquarter or half-wavelength awayfrom the drain. Since the stubs fordifferent harmonics interact and theopen or short must be created at a“virtual drain” ahead of the draincapacitance and bond-wire induc-tance, implementation of suitablenetworks is a bit of an art.Nonetheless, class-F PAs are success-fully implemented from MF throughKa band.
A variety of modes of operation in-between class C, E, and F are possi-
28 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
ble. The maximum achievable effi-ciency [30] depends upon the numberof harmonics, (0.5, 0.707, 0.8165,0.8656, 0.9045 for 1 through 5 har-monics, respectively). The utilizationfactor depends upon the harmonicimpedances and is highest for idealclass-F operation.
6b. LOAD-PULL CHARACTERIZATION
RF-power transistors are charac-terized by breakdown voltages andsaturated drain currents. The combi-nation of the resultant maximumdrain voltage and maximum draincurrent dictates a range of loadimpedances into which useful powercan be delivered, as well as animpedance for delivery of the maxi-mum power. The load impedance formaximum power results in drainvoltage and current excursions fromnear zero to nearly the maximumrated values.
The load impedances correspond-ing to delivery of a given amount ofRF power with a specified maximumdrain voltage lie along parallel-resis-
tance lines on the Smith chart. Theimpedances for a specified maximumcurrent analogously follow a series-resistance line. For an ideal PA, theresultant constant-power contour isfootball-shaped as shown in Figure11.
In a real PA, the ideal drain isembedded behind the drain capaci-tance and bond-wire/package induc-tance. Transformation of the idealdrain impedance through these ele-ments causes the constant-powercontours to become rotated and dis-torted [31]. With the addition of sec-ond-order effects, the contoursbecome elliptical. A set of power con-tours for a given PA somewhatresembles a set of contours for a con-jugate match. However, a true conju-gate match produces circular con-tours. With a power amplifier, theprocess is more correctly viewed asloading to produce a desired poweroutput. As shown in the example ofFigure 12, the power and efficiencycontours are not necessarily aligned,nor do maximum power and maxi-mum efficiency necessarily occur forthe same load impedance. Sets ofsuch “load-pull” contours are widelyused to facilitate design trade-offs.
Load-pull analyses are generallyiterative in nature, as changing one
parameter may produce a new set ofcontours. A variety of differentparameters can be plotted during aload-pull analysis, including not onlypower and efficiency, but also distor-tion and stability. Harmonicimpedances as well as driveimpedances are also sometimes var-ied.
A load-pull system consists essen-tially of a test fixture, provided withbiasing capabilities, and a pair of low-loss, accurately resettable tuners,usually of precision mechanical con-struction. A load-pull characteriza-tion procedure consists essentially ofmeasuring the power of a device, to agiven specification (e.g., the 1-dBcompression point) as a function ofimpedance. Data are measured at alarge number of impedances andplotted on a Smith chart. Such plotsare, of course, critically dependent onthe accurate calibration of the tuners,both in terms of impedance and loss-es. Such calibration is, in turn, highlydependent on the repeatability of thetuners.
Precision mechanical tuners, withmicrometer-style adjusters, were thetraditional apparatus for load-pullanalysis. More recently, a new gener-ation of electronic tuners hasemerged that tune through the usevaractors or transmission linesswitched by pin diodes. Such elec-tronic tuners [32] have the advantageof almost perfect repeatability andhigh tuning speed, but have muchhigher losses and require highly com-plex calibration routines. Mechanicaltuners are more difficult to controlusing a computer, and move veryslowly from one impedance setting toanother.
In an active load-pull system, asecond power source, synchronized infrequency and phase with the deviceinput excitation, is coupled into theoutput of the device. By controllingthe amplitude and phase of theinjected signal, a wide range ofimpedances can be simulated at theoutput of the test device [33]. Such a
Figure 11 · Contant power contoursand transformation.
Figure 12 · Example load-pull con-tours for a 0.5-W, 836 MHz PA.(Courtesy Focus Microwaves anddBm Engineering)
30 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
system eliminates the expensivetuners, but creates a substantial cali-bration challenge of its own. The wideavailability of turn-key load-pull sys-tems has generally reduced the appli-cation of active load-pull to situationswhere mechanical or electronic tun-ing becomes impractical (e.g., mil-limeter-wave frequencies).
6c. STABILITYThe stability of a small-signal RF
amplifier is ensured by deriving a setof S-parameters from using mea-sured data or a linear model, andthen establishing the value of the k-factor stability parameter. If the k-factor is greater than unity, at thefrequency and bias level in question,then expressions for matchingimpedances at input and output canbe evaluated to give a perfect conju-gate match for the device. Amplifierdesign in this context is mainly amatter of designing matching net-works which present the prescribedimpedances over the necessary speci-fied bandwidth. If the k factor is lessthan unity, negative feedback or lossymatching must be employed in orderto maintain an unconditionally stabledesign.
A third case is relevant to PAdesign at higher microwave frequen-cies. There are cases where a devicehas a very high k-factor value, butvery low gain in conjugate matchedcondition. The physical cause of thiscan be traced to a device which hasgain roll-off due to carrier-mobilityeffects, rather than parasitics. Insuch cases, introduction of some posi-tive feedback reduces the k-factorand increases the gain in conjugatelymatched conditions, while maintain-ing unconditional stability. This tech-nique was much used in the early eraof vacuum-tube electronics, especiallyin IF amplifiers.
elements (capacitors, inductors)become unsuitable as tuning compo-
nents and are used primarily aschokes and by-passes. Matching, tun-ing, and filtering at microwave fre-quencies are therefore accomplishedwith distributed (transmission-line)networks. Proper operation of poweramplifiers at microwave frequenciesis achieved by providing the requireddrain-load impedance at the funda-mental and a number of harmonicfrequencies.
Class FClass-F operation is specified in
terms of harmonic impedances, so itis relatively easy to see how trans-mission-line networks are used.Methods for using transmission linesin conjunction with lumped-elementtuned circuits appear in the originalpaper by Tyler [34]. In modernmicrowave implementation, however,it is generally necessary to use trans-mission lines exclusively. In addition,the required impedances must beproduced at a virtual ideal drain thatis separated from the output networkby drain capacitance, bond-wire/leadinductance.
Typically, a transmission linebetween the drain and the load pro-vides the fundamental-frequencydrain impedance of the desired value.A stub that is a quarter wavelengthat the harmonic of interest and openat one end provides a short circuit atthe opposite end. The stub is placedalong the main transmission line ateither a quarter or a half wavelengthfrom the drain to create either anopen or a short circuit at the drain[35]. The supply voltage is fed to thedrain through a half-wavelength linebypassed on the power-supply end oralternately by a lumped-elementchoke. When multiple stubs are used,the stub for the highest controlledharmonic is placed nearest the drain.Stubs for lower harmonics are placedprogressively further away and theirlengths and impedances are adjustedto allow for interactions. Typically,“open” means three to ten times thefundamental-frequency impedance,
and “shorted” means no more 1/10 to1/3 of the fundamental-frequencyimpedance [FR17].
A wide variety of class-F PAs havebeen implemented at UHF andmicrowave frequencies [36-41].Generally, only one or two harmonicimpedances are controlled. In the X-band PA from [42], for example, theoutput circuit provides a match at thefundamental and a short circuit atthe second harmonic. The third-har-monic impedance is high, but notexplicitly adjusted to be open. The 3-dB bandwidth of such an output net-work is about 20 percent, and the effi-ciency remains within 10 percent ofits maximum value over a bandwidthof approximately 10 to 15 percent.
Dielectric resonators can be usedin lieu of lumped-element traps inclass-F PAs. Power outputs of 40 Whave been obtained at 11 GHz withefficiencies of 77 percent [43].
Class EThe drain-shunt capacitance and
series inductive reactance requiredfor optimum class-E operation resultin a drain impedance of R + j0.725Rat the fundamental frequency,–j1.7846R at the second harmonic,and proportionately smaller capaci-tive reactances at higher harmonics.At microwave frequencies, class-Eoperation is approximated by provid-ing the drain with the fundamental-frequency impedance and preferablyone or more of the harmonicimpedances [44].
An example of a microwaveapproximation of class E that pro-vides the correct fundamental andsecond-harmonic impedances [44] isshown in Figure 13. Line l2 is a quar-ter-wavelength long at the secondharmonic so that the open circuit atits end is transformed to a short atplane AA'. Line l1 in combinationwith L and C is designed to be also aquarter wavelength to translate theshort at AA' to an open at the tran-sistor drain. The lines l1 to l4 providethe desired impedance at the funda-
July 2003 31
mental. The implementation using anFLK052 MESFET is shown in Figure14 produces 0.68 W at X band with adrain efficiency of 72 percent andPAE of 60 percent [42].
Methods exist for providing theproper impedances through thefourth harmonic [45]. However, theharmonic impedances are not critical[30], and many variations are there-fore possible. Since the transistoroften has little or no gain at the high-er harmonic frequencies, thoseimpedances often have little or noeffect upon performance. A single-stub match is often sufficient to pro-vide the desired impedance at thefundamental while simultaneouslyproviding an adequately highimpedance at the second harmonic,thus eliminating the need for anextra stub and reducing a portion ofthe losses associated with it. Mostmicrowave class-E amplifiers operatein a suboptimum mode [46].Demonstrated capabilities rangefrom 16 W with 80-percent efficiencyat UHF (LDMOS) to 100 mW with60-percent efficiency at 10 GHz [47],[48], [44], [49], [50], [51]. Optical sam-pling of the waveforms [52] has veri-fied that these PAs do indeed operatein class E.
ComparisonPAs configured for classes A (AB),
E, and F are compared experimental-ly in [50] with the following conclu-sions. Classes AB and F have essen-tially the same saturated output
power, but class F has about 15 per-cent higher efficiency. Class E has thehighest efficiency. Gain compressionoccurs at a lower power level for classE than for class F. For a given effi-ciency, class F produces more power.For the same maximum outputpower, the third order intermodula-tion products are about 10 dB lowerfor class F than for class E. Lower-power PAs implemented with smallerRF power devices tend to be moreefficient than PAs implemented withlarger devices [42].
Millimeter-Wave PAsSolid-state PAs for millimeter-
wave (mm-W) frequencies (30 to 100GHz) are predominantly monolithic.Most Ka-band PAs are based uponpHEMT devices, while most W-bandPAs are based upon InP HEMTs.Some use is also made of HBTs at thelower mm-W frequencies. Class A isused for maximum gain. Typical per-formance characteristics include 4 Wwith 30-percent PAE at Ka band [53],250 mW with 25-percent PAE at Qband [54], and 200 mW with 10-per-cent PAE at W band [55]. Devices foroperation at mm-W are inherentlysmall, so large power outputs areobtained by combining the outputs ofmultiple low-power amplifiers in cor-porate or spatial power combiners.
6e. EXAMPLE APPLICATIONSThe following examples illustrate
the wide variety of power amplifiersin use today:
HF/VHF Single SidebandOne of the first applications of
RF-power transistors was linearamplification of HF single-sidebandsignals. Many PAs developed byHelge Granberg have been widelyadapted for this purpose [56, 57]. The300-W PA for 2 to 30 MHz uses a pairof Motorola MRF422 Si NPN transis-tors in a push-pull configuration. ThePA operates in class AB push-pullfrom a 28-V supply and achieves acollector efficiency of about 45 per-cent (CW) and a two-tone IMD ratioof about –30 dBc. The 1-kW amplifieris based upon a push-pull pair ofMRF154 MOSFETs and operatesfrom a 50-V supply. Over the frequen-cy range of 2 to 50 MHz it achieves adrain efficiency of about 58 percent(CW) with an IMD rating of –30 dBc.
13.56-MHz ISM Power SourcesHigh-power signals at 13.56 MHz
are needed for a wide variety ofIndustrial, Scientific, and Medical(ISM) applications such as plasmageneration, RF heating, and semicon-ductor processing. A 400-W class-EPA uses an International RectifierIRFP450LC MOSFET (normallyused for low-frequency switching-mode DC power supplies) operatesfrom a 120-V supply and achieves adrain efficiency of 86 percent [58, 26].Industrial 13.56-MHz RF power gen-erators using class-E output stageshave been manufactured since 1992by Dressler Hochfrequenztechnik(Stolberg, Germany) and Advanced
Figure 13 · Idealized microwave class-E PA circuit. Figure 14 · Example X-band class-E PA.
32 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
Energy Industries (Ft. Collins, CO).They typically use RF-powerMOSFETs with 500- to 900-V break-down voltages made by DirectedEnergy or Advanced PowerTechnology and produce output pow-ers of 500 W to with 3 kW with drainefficiencies of about 90 percent. TheAdvanced Energy Industries amplifi-er (Figure 15) uses thick-film-hybridcircuits to reduce size. This allowsplacement inside the clean-roomfacilities of semiconductor-manufac-turing plants, eliminating the needfor long runs of coaxial cable from anRF-power generator installed outsidethe clean-room.
VHF FM Broadcast TransmitterFM-broadcast transmitters (88 to
108 MHz) with power outputs from50 W to 10 kW are manufactured byBroadcast Electronics (Quincy,Illinois). These transmitters use up to32 power-combined PAs based uponMotorola MRF151G MOSFETs. ThePAs operate in class C from a 44-Vsupply and achieve a drain efficiencyof 80 percent. Typically, about 6 per-cent of the output power is dissipatedin the power combiners, harmonic-suppression filter, and lightning-pro-tection circuit.
MF AM Broadcast TransmittersSince the 1980s, AM broadcast
transmitters (530 to 1710 kHz) havebeen made with class-D and -E RF-output stages. Amplitude modulationis produced by varying the supplyvoltage of the RF PA with a high-effi-ciency amplitude modulator.
Transmitters made by Harris(Mason, Ohio) produce peak-envelopeoutput powers of 58, 86, 150, 300, and550 kW (unmodulated carrier powersof 10, 15, 25, 50, and 100 kW). The100-kW transmitter combines theoutput power from 1152 transistors.The output stages can use eitherbipolars or MOSFETs, typically oper-ate in class DE from a 300-V supply,and achieve an efficiency of 98 per-cent. The output section of the Harris3DX50 transmitter is shown inFigure 16.
Transmitters made by BroadcastElectronics (Quincy, IL) use class-ERF-output stages based uponAPT6015LVR MOSFETs operatingfrom 130-V maximum supply volt-ages. They achieve drain efficienciesof about 94 percent with peak-enve-lope output powers from 4.4 to 44 kW.The 44-kW AM-10A transmitter com-bines outputs from 40 individual out-put stages.
900-MHz Cellular-TelephoneHandset
Most 900-MHz CDMA handsetsuse power-amplifier modules fromvendors such as Conexant and RFMicro Devices. These modules typi-cally contain a single GaAs-HBTRFIC that includes a single-endedclass-AB PA. Recently developed PAmodules also include a silicon controlIC that provides the base-bias refer-ence voltage and can be commandedto adjust the output-transistor basebias to optimize efficiency whilemaintaining acceptably low amplifierdistortion. over the full ranges oftemperature and output power. A typ-ical module (Figure 17) produces 28dBm (631 mW) at full output with aPAE of 35 to 50 percent.
Cellular-Telephone BaseStation Transmitter
The Spectrian MCPA 3060 cellu-lar base-station transmitter for 1840-1870 MHz CDMA systems providesup to 60-W output while transmittinga signal that may include as many 9modulated carriers. IMD is mini-mized by linearizing a class-AB mainamplifier with both adaptive predis-tortion and adaptive feed-forwardcancellation. The adaptive control
Figure 15 · 3-kW high efficiency PA for 13.56 ISM-band operation.(Courtesy Advanced Energy)
Figure 16 · Output section of a 50-kW AM broadcast transmitter.(Courtesy Harris)
34 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
system adjusts operation as neededto compensate for changes due totemperature, time, and output power.The required adjustments arederived from continuous measure-ments of the system response to aspread-spectrum pilot test signal.The amplifier consumes a maximumof 810 W from a 27-V supply.
S-Band Hybrid Power ModuleA thick-film-hybrid power-ampli-
fier module made by UltraRF (nowCree Microwave) for 1805 to 1880MHz DCS and 1930-1960 MHz PCSis shown in Figure 18. It uses four140-mm LDMOS FETs operatingfrom a 26-V drain supply. The indi-vidual PAs have 11-dB power gainand are quadrature-combined to pro-duce a 100-W PEP output. The aver-age output power is 40 W for EDGEand 7 W for CDMA, with an ACPR of–57 dBc for EDGE and –45 dBc forCDMA. The construction is basedupon 0.02-in. thick film with silvermetalization.
GaAs MMIC Power AmplifierA MMIC PA for use from 8 to 14
GHz is shown in Figure 19. Thisamplifier is fabricated with GaAsHBTs and intended for used inphased-array radar. It produces a 3-W output with a PAE of approxi-mately 40 percent [59].
References19. H. L. Krauss, C. W. Bostian, and
F. H. Raab, Solid State RadioEngineering, New York: Wiley, 1980.
20. R. Gupta and D. J. Allstot, “Fullymonolithic CMOS RF power amplifiers:Recent advances,” IEEE Communi-cations Mag., vol. 37, no. 4, pp. 94-98,April 1999.
21. F. H. Raab and D. J. Rupp, “HFpower amplifier operates in both classB and class D,” Proc. RF Expo West ’93,San Jose, CA, pp. 114-124, March 17-19,1993.
22. P. Asbeck, J. Mink, T. Itoh, and G.Haddad, “Device and circuit approaches
23. N. O. Sokal and A. D. Sokal,“Class E—a new class of high efficiencytuned single-ended switching poweramplifiers,” IEEE J. Solid-StateCircuits, vol. SC-10, no. 3, pp. 168-176,June 1975.
24. F. H. Raab, “Effects of circuitvariations on the class E tuned poweramplifier,” IEEE J. Solid State Circuits,vol. SC-13, no. 2, pp. 239-247, April1978.
25. F. H. Raab, “Effects of VSWRupon the class-E RF-power amplifier,”Proc. RF Expo East ’88, Philadelphia,PA, pp. 299-309, Oct. 25-27, 1988.
26. J. F. Davis and D. B. Rutledge, “Alow-cost class-E power amplifier withsine-wave drive,” Int. Microwave Symp.Digest, vol. 2, pp. 1113-1116, Baltimore,MD, June 7-11, 1998.
27. T. B. Mader and Z. B. Popovic,“The transmission-line high-efficiencyclass-E amplifier,” IEEE Microwaveand Guided Wave Letters, vol. 5, no. 9,pp. 290-292, Sept. 1995.
28. D. C. Hamill, “Class DE invert-ers and rectifiers for DC-DC conver-sion,” PESC96 Record, vol. 1, pp. 854-860, June 1996.
29. F. H. Raab, “Maximum efficiencyand output of class-F power amplifiers,”IEEE Trans. Microwave Theory Tech.,vol. 47, no. 6, pp. 1162-1166, June 2001.
30. F. H. Raab, “Class-E, -C, and -Fpower amplifiers based upon a finitenumber of harmonics,” IEEE Trans.Microwave Theory Tech., vol. 47, no. 8,pp. 1462-1468, Aug. 2001.
31. S. C. Cripps, RF PowerAmplifiers for Wireless Communi-cation, Norwood, MA: Artech, 1999.
32. “A load pull system with har-monic tuning,” Microwave J., pp. 128-132, March 1986.
33. B. Hughes, A. Ferrero, and A.Cognata, “Accurate on-wafer power andharmonic measurements of microwaveamplifiers and devices,” IEEE Int.Microwave Symp. Digest, Albuquerque,NM, pp. 1019-1022, June 1-5, 1992.
34. V. J. Tyler, “A new high-efficiency
Figure 17 · Internal view of a dual-band (GSM/DCS) PA module forcellular telephone handsets.(Courtesy RF Micro Devices)
high power amplifier,” The MarconiReview, vol. 21, no. 130, pp. 96-109, Fall1958.
35. A. V. Grebennikov, “Circuitdesign technique for high-efficiencyclass-F amplifiers,” Int. MicrowaveSymp. Digest, vol. 2, pp. 771-774,Boston, MA, June 13-15, 2000.
36. P. Colantonio, F. Giannini, G.Leuzzi, and E. Limiti, “On the class-Fpower-amplifier design,” RF and Micro-wave Computer-Aided Engin-eering,vol. 32, no. 2, pp. 129-149, March 1999.
37. A. N. Rudiakova and V. G.Krizhanovski, “Driving waveforms forclass-F power amplifiers,” Int.Microwave Symp. Digest, vol. 1, pp. 473-476, Boston, MA, June 13-15, 2000.
38. A. Inoue, T. Heima, A. Ohta, R.Hattori, and Y. Mitsui, “Analysis ofclass-F and inverse class-F amplifiers,”Int. Microwave Symp. Digest, vol. 2, pp.775-778, Boston, MA, June 13-15, 2000.
39. F. van Rijs et al., “Influence ofoutput impedance on power added effi-ciency of Si-bipolar power transistors,”Int. Microwave Symp. Digest, vol. 3, pp.1945-1948, Boston, MA, June 13-15,2000.
40. F. Huin, C. Duvanaud, V. Serru,F. Robin, and E. Leclerc, “A single supplyvery high power and efficiency integrat-ed PHEMT amplifier for GSM applica-tions,” Proc. 2000 RFIC Symp., Boston,MA, CD-ROM, June 11-13, 2000.
41. B. Ingruber et al.,“Rectangularly driven class-A harmon-ic-control amplifier,” IEEE Trans.Microwave Theory Tech., pt. 1, vol. 46,no. 11, pp. 1667-1672, Nov. 1998.
42. E. W. Bryerton, M. D. Weiss, andZ. Popovic, “Efficiency of chip-level ver-sus external power combining,” IEEETrans. Microwave Theory Tech., vol. 47,no. 8, pp. 1482-1485, Aug. 1999.
43 S. Toyoda, “Push-pull poweramplifiers in the X band,” Int.Microwave Symp. Digest, vol. 3, pp.1433-1436, Denver, CO, June 8-13, 1997.
44. T. B. Mader and Z. Popovic, “Thetransmission-line high-efficiency class-E amplifier,” IEEE Micro-wave andGuided Wave Lett., vol. 5, no. 9, pp. 290-292, Sept. 1995.
45. A. J. Wilkinson and J. K. A.Everard, “Transmission line load net-work topology for class E amplifiers,”IEEE Trans. Microwave Theory Tech.,vol. 47, no. 6, pp. 1202-1210, June 2001.
46. F. H. Raab, “Suboptimum opera-tion of class-E power amplifiers,” Proc.RF Technology Expo., Santa Clara, CA,pp. 85-98, Feb. 1989.
47. S. Li, “UHF and X-band class-Eamplifiers,” Ph.D. Thesis, CaliforniaInstitute of Technology, Pasadena, 1999.
48. F. J. Ortega-Gonzalez, J. L.Jimenez-Martin, A. Asensio-Lopez, G.Torregrosa-Penalva, “High-efficiencyload-pull harmonic controled class-Epower amplifier,” IEEE MicrowaveGuided Wave Lett., vol. 8, no. 10, pp.348-350, Oct. 1998.
49. E. Bryerton, “High-efficiencyswitched-mode microwave circuits,”Ph.D. dissertation, Univ. of Colorado,Boulder, June 1999.
50. T. B. Mader, E. W. Bryerton, M.Markovic, M. Forman, and Z. Popovic,“Switched-mode high-efficiency micro-wave power amplifiers in a free-spacepower-combiner array,” IEEE Trans.Microwave Theory Tech., vol. 46, no. 10,pt. I, pp. 1391-1398, Oct. 1998.
51. M. D. Weiss and Z. Popovic, “A 10GHz high-efficiency active antenna,”Int. Microwave Symp. Digest, vol. 2, pp.663-666, Anaheim, CA, June 14-17,1999.
52. M. Weiss, M. Crites, E. Bryerton,J. Whitacker, and Z. Popovic, “"Timedomain optical sampling of nonlinearmicrowave amplifiers and multipliers,”IEEE Trans. Microwave Theory Tech.,vol. 47, no.12, pp. 2599-2604, Dec. 1999.
53. J. J. Komiak, W. Kong, P. C.Chao, and K. Nichols, “Fully monolithic4 watt high efficiency Ka-band poweramplifier,” Int. Microwave Symp.Digest, vol. 3, pp. 947-950, Anaheim,CA, June 14-17, 1999.
54. S.-W. Chen et al., “A 60-GHzhigh-efficiency monolithic power ampli-fier using 0.1-µm pHEMTs,” IEEEMicrowave and Guided Wave Lett.,vol.5, pp. 201-203, June 1995.
55. D. L. Ingram et al., “Compact W-band solid-state MMIC high power
56. H. Granberg, “Get 300 wattsPEP linear across 2 to 30 MHz fromthis push-pull amplifier,” BulletinEB27A, Motorola SemiconductorProducts, Phoenix, Feb. 1980.
57. H. Granberg, “A compact 1-kW2-50 MHz solid-state linear amplifier,”QEX, no. 101, pp. 3-8, July 1990. AlsoAR347, Motorola SemiconductorProducts, Feb. Oct. 1990.
58. N. O. Sokal, “Class-E RF poweramplifiers ... ,” QEX, No. 204, pp. 9-20,Jan./Feb. 2001.
59. M. Salib, A. Gupta, A. Ezis, M.Lee, and M. Murphy, “A robust 3W highefficiency 8-14 GHz GaAs/AlGaAs het-erojunction bipolar transistor poweramplifier,” Int. Microwave Symp.Digest, vol. 2, pp. 581-584, Baltimore,MD, June 7-11, 1998.
Author InformationThe authors of this series of arti-
cles are: Frederick H. Raab (leadauthor), Green Mountain RadioResearch, e-mail: [email protected];Peter Asbeck, University ofCalifornia at San Diego; SteveCripps, Hywave Associates; Peter B.Kenington, Andrew Corporation;Zoya B. Popovic, University ofColorado; Nick Pothecary,Consultant; John F. Sevic, CaliforniaEastern Laboratories; and Nathan O.Sokal, Design Automation. Readersdesiring more information shouldcontact the lead author.
Notes1. In Part 1 of this series (May
2003 issue), the references containedin Table 1 were not numbered cor-rectly. The archived version has beencorrected and may be downloadedfrom: www.highfrequencyelectronics.com — click on “Archives,” select“May 2003 — Vol. 2 No. 3” then clickon the article title.
2. This series has been extendedto five parts, to be published in succe-sive issues through January 2004.
34 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
RF and Microwave PowerAmplifier and TransmitterTechnologies — Part 3
By Frederick H. Raab, Peter Asbeck, Steve Cripps, Peter B. Kenington,Zoya B. Popovich, Nick Pothecary, John F. Sevic and Nathan O. Sokal
The building blocksused in transmit-ters are not only
power amplifiers, but avariety of other circuitelements including oscil-lators, mixers, low-levelamplifiers, filters, match-ing networks, combiners,and circulators. The
arrangement of building blocks is known asthe architecture of a transmitter. The classictransmitter architecture is based upon linearPAs and power combiners. More recently,transmitters are being based upon a variety ofdifferent architectures including stagebypassing, Kahn, envelope tracking, outphas-ing, and Doherty. Many of these are actuallyfairly old techniques that have been recentlymade practical by the capabilities of DSP.
7a. LINEAR ARCHITECTUREThe conventional architecture for a linear
microwave transmitter consists of a basebandor IF modulator, an up-converter, and a power-amplifier chain (Figure 20). The amplifierchain consists of cascaded gain stages withpower gains in the range of 6 to 20 dB. If thetransmitter must produce an amplitude-mod-ulated or multi-carrier signal, each stage musthave adequate linearity. This generallyrequires class-A amplifiers with substantialpower back-off for all of the driver stages. Thefinal amplifier (output stage) is always themost costly in terms of device size and currentconsumption, hence it is desirable to operatethe output stage in class B. In applicationsrequiring very high linearity, it is necessary touse class A in spite of the lower efficiency.
The outputs of a driver stage must bematched to the input of the following stagemuch as the final amplifier is matched to theload. The matching tolerance for maintainingpower level can be significantly lower thanthat for gain [60], hence the 1-dB load-pullcontours are more tightly packed for powerthan for gain.
To obtain even modest bandwidths (e.g.,above 5 percent), the use of quadrature bal-anced stages is advisable (Figure 21). Themain benefit of the quadrature balanced con-figuration is that reflections from the transis-tors are cancelled by the action of the inputand output couplers. An individual device cantherefore be deliberately mismatched (e.g., toachieve a power match on the output), yet thequadrature-combined system appears to bewell-matched. This configuration also acts asan effective power combiner, so that a givenpower rating can be achieved using a pair ofdevices having half of the required power per-formance. For moderate-bandwidth designs,the lower-power stages are typically designedusing a simple single-ended cascade, which insome cases is available as an RFIC. Designswith bandwidths approaching an octave or
Transmitter architectures isthe subject of this install-
more require the use of quadrature-balanced stages throughout theentire chain.
Simple linear-amplifier chains ofthis kind have high linearity but onlymodest efficiency. Single-carrierapplications usually operate the finalamplifier to about the 1-dB compres-sion point on amplitude modulationpeaks. A thus-designed chain inwhich only the output stage exhibitscompression can still deliver anACPR in the range of about –25 dBcwith 50-percent efficiency at PEP.
Two practical problems are fre-quently encountered in the design oflinear PA chains: stability and lowgain. Linear, class-A chains are actu-ally more susceptible to oscillationdue to their high gain, and single-path chains are especially prone tounstable behavior. Instability can besubdivided into the two distinct cate-gories: Low-frequency oscillation andin-band instability. In-band instabili-ty is avoided by designing the indi-vidual gain stages to meet the crite-ria for unconditional stability; i.e.,the Rollet k factor [61] must begreater than unity for both in-bandand out-of-band frequencies. Meetingthis criterion usually requires sacri-ficing some gain through the use ofabsorptive elements. Alternatively,the use of quadrature balancedstages provides much greater isola-tion between individual stages, andthe broadband response of thequadrature couplers can eliminatethe need to design the transistor
stage itself with k>1. This is anotherreason for using quadrature coupledstages in the output of the chain.
Large RF power devices typicallyhave very high transconductance, andthis can produce low-frequency insta-bility unless great care is taken toterminate both the input and outputat low frequencies with impedancesfor unconditional stability. Because oflarge separation from the RF band,this is usually a simple matter requir-ing a few resistors and capacitors.
At X band and higher, the powergain of devices in the 10 W and abovecategory can drop well below 10 dB.To maintain linearity, it may be nec-essary to use a similarly size deviceas a driver. Such an architectureclearly has a major negative impactupon the cost and efficiency of thewhole chain. In the more extremecases, it may be advantageous to con-sider a multi-way power combiner,where 4, 8, or an even greater num-ber of smaller devices are combined.Such an approach also has otheradvantages, such as soft failure, bet-ter thermal management, and phaselinearity. However, it typically con-sumes more board space.
7b. POWER COMBINERSThe need frequently arises to
combine the outputs of several indi-vidual PAs to achieve the desiredtransmitter output. Whether to use anumber of smaller PAs vs. a singlelarger PA is one of the most basicdecisions in selection of an architec-
ture [60]. Even when larger devicesare available, smaller devices oftenoffer higher gain, a lower matching Qfactor (wider bandwidth), betterphase linearity, and lower cost. Heatdissipation is more readily accom-plished with a number of smalldevices, and a soft-failure modebecomes possible. On the other hand,the increase in parts count, assemblytime, and physical size are significantdisadvantages to the use of multiple,smaller devices.
Direct connection of multiple PAsis generally impractical as the PAsinteract, allowing changes in outputfrom one to cause the load impedanceseen by the other to vary. A constantload impedance, hence isolation ofone PA from the other, is provided bya hybrid combiner. A hybrid combinercauses the difference between thetwo PA outputs to be routed to anddissipated in a balancing or “dump”resistor. In the event that one PAfails, the other continues to operatenormally, with the transmitter out-put reduced to one fourth of nominal.
The most common power combin-er is the quadrature-hybrid combiner.A 90° phase shift is introduced atinput of one PA and also at the out-put of the other. The benefits ofquadrature combining include con-stant input impedance in spite ofvariations of input impedances of theindividual PAs, cancellation of oddharmonics, and cancellation of back-ward-IMD (IMD resulting from a sig-nal entering the output port). Inaddition, the effect of load impedanceupon the system output is greatlyreduced (e.g., to 1.2 dB for a 3:1SWR). Maintenance of a nearly con-stant output occurs because the loadimpedance presented to one PAdecreases when that presented to theother PA increases. As a result, how-ever, device ratings increase and effi-ciency decreases roughly in propor-tion to the SWR [65]. Becausequadrature combiners are inherentlytwo-terminal devices, they are usedin a corporate combining architecture
Figure 21 · Amplifier stages with quadrature combiners.
0º
90º
38 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
(Figure 21). Unfortunately, the physical construction ofsuch couplers poses some problems in a PC-board envi-ronment. The very tight coupling between the two quar-ter-wave transmission lines requires either very fine gapsor a three-dimensional structure. This problem is circum-vented by the use of a miniature co-axial cable having apair of precisely twisted wires to from the coupling sec-tion or ready-made, low-cost surface mount 3-dB couplers.
The Wilkinson or in-phase power combiner [62] isoften more easily fabricated than a quadrature combiner.In the two-input form (as in each section in Figure 22),the outputs from two quarter-wavelength lines summedinto load R0 produce an apparent load impedance of 2R0,which is transformed through the lines into at the loadimpedances RPA seen by the individual PAs. The differ-ence between the two PA outputs is dissipated in a resis-tor connected across the two inputs. Proper choice of thebalancing resistor (2RPA) produces a hybrid combinerwith good isolation between the two PAs. The Wilkinsonconcept can be extended to include more than two inputs[63].
Greater bandwidth can be obtained by increasing thenumber of transforming sections in each signal path. Asingle-section combiner can have a useful bandwidth ofabout 20 percent, whereas a two-section version can havea bandwidth close to an octave. In practice, escalating cir-cuit losses generally preclude the use of more than twosections.
All power-combining techniques all suffer from circuitlosses as well as mismatch losses. The losses in a simpletwo-way combiner are typically about 0.5 dB or 10 per-cent. For a four-way corporate structure, the intercon-nects typically result in higher losses. Simple openmicrostrip lines are too lossy for use in combining struc-tures. One technique that offers a good compromiseamong cost, produceability, and performance, uses sus-pended stripline. The conductors are etched onto double-sided PC board, interconnected by vias, and then sus-
pended in a machined cavity. Structures of this kind allowhigh-power 8-way combiners with octave bandwidths andof 0.5 dB.
A wide variety of other approaches to power-combin-ing circuits are possible [62, 64]. Microwave power canalso be combined during radiation from multiple anten-nas through “quasi-optical” techniques [66].
7c. STAGE SWITCHING AND BYPASSINGThe power amplifier in a portable transmitter gener-
ally operates well below PEP output, as discussed inSection 4 (Part 1). The size of the transistor, quiescentcurrent, and supply voltage are, however, determined bythe peak output of the PA. Consequently, a PA with alower peak output produces low-amplitude signals moreefficiently than does a PA with a larger peak output, asillustrated in Figure 23 for class-B PAs with PEP effi-ciencies of 60 percent. Stage-bypassing and gate-switch-ing techniques [67, 68] reduce power consumption andincrease efficiency by switching between large and smallamplifiers according to signal level. This process is analo-gous to selection of supply voltage in a class-G PA, andthe average efficiency can be similarly computed [69].
A typical stage-bypassing architecture is shown inFigure 24. For low-power operation, switches SA and SBroute the drive signal around the final amplifier.
Figure 22 · Multi-section Wilkinson combining architecture.Figure 23 · Power consumption byPAs of different sizes.
Figure 24 · Stage-bypassing architecture.
40 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
Simultaneously, switch SDC turns-offDC power to the final amplifier. Thereduction in power consumption canimprove the average efficiency signif-icantly (e.g., from 2.1 to 9.5 percent in[70]). The control signal is based uponthe signal envelope and power level(back-off). Avoiding hysteresis effectsand distortion due to switching tran-sients are critical issues in imple-mentation.
A PA with adaptive gate switchingis shown in Figure 25. The gate width(hence current and power capability)of the upper FET is typically ten totwenty times that of the lower FET.The gate bias for the high-power FETkeeps it turned off unless it is neededto support a high-power output.Consequently, the quiescent draincurrent is reduced to low levels unlessactually needed. The advantages ofthis technique are the absence of lossin the switches required by stagebypassing, and operation of the low-power FET in a more linear region(vs. varying the gate bias of a singlelarge FET). The disadvantage is thatthe source and load impedanceschange as the upper FET is switchedon and off.
7d. KAHN TECHNIQUEThe Kahn Envelope Elimination
and Restoration (EER) technique(Figure 26) combines a highly effi-
cient but nonlinear RF power amplifi-er (PA) with a highly efficient enve-lope amplifier to implement a high-efficiency linear RF power amplifier.In its classic form [73], a limiter elim-inates the envelope, allowing the con-stant-amplitude phase modulatedcarrier to be amplified efficiently byclass-C, -D, -E, or -F RF PAs.Amplitude modulation of the final RFPA restores the envelope to the phase-modulated carrier creating an ampli-fied replica of the input signal.
EER is based upon the equiva-lence of any narrowband signal tosimultaneous amplitude (envelope)and phase modulations. In a modernimplementation, both the envelopeand the phase-modulated carrier aregenerated by a DSP. In contrast tolinear amplifiers, a Kahn-techniquetransmitter operates with high effi-ciency over a wide dynamic rangeand therefore produces a high aver-age efficiency for a wide range of sig-nals and power (back-off) levels.Average efficiencies three to fivetimes those of linear amplifiers havebeen demonstrated (Figure 27) fromHF [74] to L band [75].
Transmitters based upon theKahn technique generally have excel-lent linearity because linearitydepends upon the modulator ratherthan RF power transistors. The twomost important factors affecting the
linearity are the envelope bandwidthand alignment of the envelope andphase modulations. As a rule ofthumb, the envelope bandwidth mustbe at least twice the RF bandwidthand the misalignment must notexceed one tenth of the inverse of theRF bandwidth [76]. In practice, thedrive is not hard-limited as in theclassical implementation. Drivepower is conserved by allowing thedrive to follow the envelope except atlow levels. The use of a minimumdrive level ensures proper operationof the RF PA at low signal levelswhere the gain is low [77]. At highermicrowave frequencies, the RF powerdevices exhibit softer saturationcharacteristics and larger amounts ofamplitude-to-phase conversion,necessitating the use of predistortionfor good linearity [78].
Figure 27 · Efficiency of Kahn-tec-nique transmitters.
42 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
Class-S ModulatorA class-S modulator (Figure 28) uses a transistor and
diode or a pair of transistors act as a two-pole switch togenerate a rectangular waveform with a switching fre-quency several times that of the output signal. The widthof pulses is varied in proportion to the instantaneousamplitude of the desired output signal, which is recoveredby a low-pass filter. Class S is ideally 100 percent efficientand in practice can have high efficiency over a widedynamic range. Class-S modulators are typically used asparts of a Kahn-technique transmitter, while class-Samplifiers are becoming popular for the efficient produc-tion of audio power in portable equipment. A class-S mod-ulator can be driven by a digital (on/off) signal supplieddirectly from a DSP, eliminating the need for intermedi-ate conversion to an analog signal.
Selection of the output filter is a compromise betweenpassing the infinite-bandwidth envelope and rejectingFM-like spurious components that are inherent in thePWM process. Typically, the switching frequency must besix to seven times the RF bandwidth. Modulators withswitching frequencies of 500 kHz are readily implement-ed using discrete MOSFETs and off-the-shelf ICs [74],while several MHz can be achieved using MOS ASICs ordiscrete GaAs devices [75].
Class-G ModulatorA class-G modulator (Figure 29) is a combination of lin-
ear series-pass (class-B) amplifiers that operate from dif-ferent supply voltages. Power is conserved by selecting theone with the lowest useable supply voltage [69] so that thevoltage drop across the active device is minimized.
Split-Band ModulatorMost of the power in the envelope resides at lower fre-
quencies; typically 80 percent is in the DC component.The bandwidth of a class-S modulator can therefore beextended by combining it with a linear amplifier. Whilethere are a number of approaches, the highest efficiency
(typically 90 percent) is achieved by a diplexing combiner.Obtaining a flat frequency response and resistive loadsfor the two PAs is achieved by splitting the input signalsin a DSP that acts as a pair of negative-component filters(Figure 30) [79]. The split-band modulator should makepossible Kahn-technique transmitters with RF band-widths of tens or even hundreds of MHz.
7e. ENVELOPE TRACKINGThe envelope-tracking architecture (Figure 31) is sim-
ilar to that of the Kahn technique. However, the finalamplifier operates in a linear mode and the supply volt-age is varied dynamically to conserve power [81, 82]. TheRF drive contains both amplitude and phase information,and the burden of providing linear amplification liesentirely on the final RF PA. The role of the variable powersupply is only to optimize efficiency.
Typically, the envelope is detected and used to controla DC-DC converter. While both buck (step-down) or boost(step-up) converters are used, the latter is more commonas it allows operation of the RF PA from a supply voltagehigher than the DC-supply voltage. This configuration is
also more amenable to the use of npn or n-channel tran-sistors for fast switching. The result is a minimum VDDRFcorresponding to the DC-supply voltage and tracking oflarger envelopes with a fixed “headroom” to ensure linearoperation of the RF PA. If the RF PA is operated in classA, its quiescent current can also be varied.
In general, excess power-supply voltage translates toreduced efficiency, rather than output distortion. In prin-ciple, perfect tracking of the envelope by the supply volt-age preserves the peak efficiency of the RF PA for all out-put amplitudes, as in the Kahn technique. In practice,efficiency improvement is obtained over a limited range ofoutput power.
A high switching frequency in the DC-DC converterallows both a high modulation bandwidth and the use ofsmaller inductors and capacitors. The switching devicesin the converter can in fact be implemented using thesame same transistor technology used in the RF PA.Converters with switching frequencies of 10 to 20 MHzhave recently been implemented using MOS ASICs [80],GaAs HBTs [83, 84] and RF power MOSFETs [85].
Representative results for an envelope-tracking trans-mitter based on a GaAs FET power amplifier are shown inFigure 32. The efficiency is lower at high power than thatof the conventional amplifier with constant supply voltagedue to the inefficiency of the DC-DC converter. However,the efficiency is much higher over a wide range of outputpower, with the average efficiency approximately 40 per-cent higher than that of the conventional linear amplifier.
Spurious outputs can be produced by supply-voltageripple at the switching frequency. The effects of the ripplecan be minimized by making the switching frequency suf-ficiently high or by using an appropriate filter. Variationof the RF PA gain with supply voltage can introduce dis-tortion. Such distortion can, however, be countered by pre-distortion techniques [to be covered in Section 8 (Part 4)].
7f. OUTPHASINGOutphasing was invented during the 1930s as a
means of obtaining high-quality AM from vacuum tubeswith poor linearity [86] and was used through about 1970in RCA “Ampliphase” AM-broadcast transmitters. In the1970s, it came into use at microwave frequencies underthe name “LINC” (Linear Amplification using NonlinearComponents) [87].
An outphasing transmitter (Figure 33) produces anamplitude-modulated signal by combining the outputs oftwo PAs driven with signals of different time-varyingphases. Basically, the phase modulation causes theinstantaneous vector sum of the two PA outputs to followthe desired signal amplitude (Figure 34). The inversesine of envelope E phase-modulates the driving signalsfor the two PAs to produce a transmitter output that isproportional to E. In a modern implementation, a DSPand synthesizer produce the inverse-sine modulations ofthe driving signals.
Hybrid combining (Figure 33) isolates the PAs fromthe reactive loads inherent in outphasing, allowing themto see resistive loads at all signal levels. However, bothPAs deliver full power all of the time. Consequently, theefficiency of a hybrid-coupled outphasing transmittervaries with the output power (Figure 35), resulting in anaverage efficiency that is inversely proportional to peak-to-average ratio (as for class A). Recovery of the powerfrom the dump port of the hybrid combiner offers someimprovement in the efficiency [88].
The phase of the output current is that of the vectorsum of the two PA-output voltages. Direct summation ofthe out-of-phase signals in a nonhybrid combiner inher-ently results in reactive load impedances for the poweramplifiers [89]. If the reactances are not partially can-celled as in the Chireix technique, the current drawn fromthe PAs is proportional to the transmitter-output voltage.
Figure 31 · Envelope-tracking architecture.Figure 32 · Efficiency of a GaAs FET envelope-tracking transmitter.
High Frequency Design
RF POWER AMPLIFIERS
This results in an efficiency charac-teristic similar to that of a class-B PA.
The Chireix technique [86] usesshunt reactances on the inputs to thecombiner (Figure 36) to tune-out thedrain reactances at a particularamplitude, which in turn maximizesthe efficiency in the vicinity of thatamplitude. The efficiency at high andlow amplitudes may be degraded. Inthe classic Chireix implementation,the shunt reactances maximize theefficiency at the level of the unmodu-lated carrier in an AM signal and pro-duce good efficiency over the upper 6dB of the output range. With judi-cious choice of the shunt suscep-tances, the average efficiency can bemaximized for any given signal [89,90]. For example, a normalized sus-ceptance of 0.11 peaks the instanta-neous efficiency at a somewhat loweramplitude, resulting in an averageefficiency of 52.1 percent for an idealclass-B PA and a 10-dB Rayleigh-envelope signal (vs. 28 percent for lin-
ear amplification).Virtually all microwave outphas-
ing systems in use today are of thehybrid-coupled type. Use of theChireix technique at microwave fre-quencies is difficult becausemicrowave PAs do not behave as idealvoltage sources. Simulations suggestthat direct (nonhybrid) combiningincreases both efficiency and distor-tion [91]. Since outphasing offers awide bandwidth and the distortioncan be mitigated by techniques suchas predistortion, directly coupled andChireix techniques should be fruitfulareas for future investigation.
7g. DOHERTY TECHNIQUEDevelopment of the Doherty tech-
nique in 1936 [92] was motivated bythe observation that signals with sig-nificant amplitude modulationresulted in low average efficiency.The classical Doherty architecture(Figure 37) combines two PAs ofequal capacity through quarter-wave-
length lines or networks. The “carri-er” (main) PA is biased in class Bwhile the “peaking” (auxiliary) PA isbiased in class C. Only the carrier PAis active when the signal amplitude ishalf or less of the PEP amplitude.Both PAs contribute output powerwhen the signal amplitude is largerthan half of the PEP amplitude
Operation of the Doherty systemcan be understood by dividing it intolow-power, medium-power (load-mod-ulation), and peak-power regions[96]. The current and voltage rela-tionships are shown in Figure 38 forideal transistors and lossless match-ing networks. In the low-powerregion, the instantaneous amplitudeof the input signal is insufficient toovercome the class-C (negative) biasof the peaking PA, thus the peakingPA remains cut-off and appears as anopen-circuit. With the example loadimpedances shown in Figure 37, thecarrier PA sees a 100 ohm load andoperates as an ordinary class-Bamplifier. The drain voltage increaseslinearly with output until reachingsupply voltage VDD. The instanta-neous efficiency at this point (–6 dBfrom PEP) is therefore the 78.5 per-cent of the ideal class-B PA.
As the signal amplitude increasesinto the medium-power region, thecarrier PA saturates and the peakingPA becomes active. The additionalcurrent I2 sent to the load by thepeaking PA causes the apparent loadimpedance at VL to increase above
Figure 35 · Efficiency of outphasingtransmitters with ideal class-B PAs.
44 High Frequency Electronics
46 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
the 25 ohms of the low-power region.Transformation through the quarter-wavelength line results in a decreasein the load presented to the carrierPA. The carrier PA remains in satu-ration and acts as a voltage source. It
operates at peakefficiency, butdelivers an in-creasing amountof power. At PEPoutput, both PAssee 50-ohm loadsand each delivershalf of the systemoutput power. ThePEP efficiency isthat of the class-BPAs.
The resultinginstantaneous-efficiency curve isshown in Figure
39. The classical power division (α =0.5) approximately maximizes theaverage efficiency for full-carrier AMsignals, as well as modern single-car-rier digital signals. The use of otherpower-division ratios allows the lower
efficiency peak to be shifted leftwardso that the average efficiency isincreased for signals with higherpeak-to-average ratios. For example,α = 0.36 results in a 60 percent aver-age efficiency for a Rayleigh-envelopesignal with a 10-dB peak-to-averageratio, which is a factor of 2.1 improve-ment over class B. Doherty transmit-ters with unequal power division canbe implemented by using differentPEP load impedances and differentsupply voltages in the two PAs [97].
Much recent effort has focused onaccommodating non-ideal effects(e.g., nonlinearity, loss, phase shift)into a Doherty architecture [93, 94,95]. The power consumed by the qui-escent current of the peaking amplifi-er is also a concern. The measuredACPR characteristics of an S-bandDoherty transmitter are compared tothose of quadrature-combined class-B PAs in Figure 40. The signal is IS-95 forward link with pilot channel,paging channel, and sync-channel.The PAs are based upon 50-WLDMOS transistors. Back-off is var-ied to trade-off linearity against out-put. For the specified ACPR of –45dBc, the average PAE is nearly twicethat of the quadrature-combined PAs.
In a modern implementation, DSPcan be used to control the drive andbias to the two PAs, for precise con-trol and higher linearity. It is alsopossible to use three or more stagesto keep the instantaneous efficiencyrelatively high over a larger dynamicrange [96, 98]. For ideal class-B PAs,the average efficiency of a three-stageDoherty can be as high as 70 percentfor a Rayleigh-envelope signal with10-dB peak-to-average ratio.
References60. S. C. Cripps, RF Power Amplifiers
for Wireless Communication, Norwood,MA: Artech, 1999.
61. J. M. Rollett, “Stability and power-gain invariants of linear twoports,” IRETrans. Circuit Theory, pp, 29-32, March1962.
62. S. Cohn, “A class of broadbandthree-port TEM modes hybrids,” IEEETrans. Microwave Theory Tech., vol. MTT-
Figure 37 · Doherty transmitter.
Figure 38 · Ideal voltage and current relationships in Doherty transmitter.
Figure 39 Instantaneous efficiencyof the Doherty system with idealclass-B PAs.
Figure 40 · Measured ACPR perfor-mance of an S-band Doherty trans-mitter.
48 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
16, no. 2, pp. 110-116, 1968.63. J. Goel, “K band GaAsFET amplifi-
er with 8.2 W output using a radial powercombiner,” IEEE Trans. Microwave TheoryTech., vol. MTT-32, no. 3, pp. 317-324,1984.
64. A. Berr and D. Kaminsky, “Thetravelling wave power divider/combiner,”IEEE Trans. Microwave Theory Tech., vol.MTT-28, no.12, pp. 1468-1473, 1980.
65. F. H. Raab, “Hybrid and quadra-ture splitters and combiners,” ResearchNote RN97-22, Green Mountain RadioResearch Company, Colchester, VT, Sept.20, 1997.
66. R. A. York and Z. B. Popovic, Activeand Quasi-Optical Arrays for Solid-StatePower Combining, New York: Wiley, 1997.
67. S. Brozovich, “High efficiency mul-tiple power level amplifier circuit,” U.S.Patent 5,661,434, Aug. 1997.
68. J. Sevic, “Efficient parallel stageamplifier,” U.S. Patent 5,872,481, Feb.1999.
69. F. H. Raab, “Average efficiency ofClass-G power amplifiers,” IEEE Trans.Consumer Electronics, vol. CE-32, no. 2,pp. 145-150, May 1986.
70. J. Staudinger, “Applying switchedgain stage concepts to improve efficiencyand linearity for mobile CDMA poweramplification,” Microwave J., vol. 43, no. 9,pp. 152-162, Sept. 2000.
71. “30 way radial combiner for minia-ture GaAsFET power applications”
72. T. C. Choinski, “Unequal powerdivision using several couplers to splitand recombine the output,” IEEE Trans.Microwave Theory Tech., vol. MTT-32,no.6, pp. 613-620, 1984.
73. L. R. Kahn, “Single sidebandtransmission by envelope elimination andrestoration,” Proc. IRE, vol. 40, no. 7, pp.803-806, July 1952.
74. F. H. Raab and D. J. Rupp, “High-efficiency single-sideband HF/ VHF trans-mitter based upon envelope eliminationand restoration,” Proc. Sixth Int. Conf. HFRadio Systems and Techniques (HF ’94)(IEE CP 392), York, UK, pp. 21-25, July 4- 7, 1994.
75. F. H. Raab, B. E. Sigmon, R. G.Myers, and R. M. Jackson, “L-band trans-mitter using Kahn EER technique,” IEEETrans. Microwave Theory Tech., pt. 2, vol.46, no. 12, pp. 2220-2225, Dec. 1998.
76. F. H. Raab, “Intermodulation dis-tortion in Kahn-technique transmitters,”IEEE Trans. Microwave Theory Tech., vol.44, no. 12, part 1, pp. 2273-2278, Dec.1996.
77. F. H. Raab, “Drive modulation inKahn-technique transmitters,” Int.Microwave Symp. Digest, vol. 2, pp. 811-
814, Anaheim, CA, June 1999.78. M. D. Weiss, F. H. Raab, and Z. B.
Popovic, “Linearity characteristics of X-band power amplifiers in high-efficiencytransmitters,” IEEE Trans. MicrowaveTheory Tech., vol. 47, no. 6, pp. 1174-1179,June 2001.
79. F. H. Raab, “Technique for wide-band operation of power amplifiers andcombiners,” U. S. Patent 6,252,461, June26, 2001.
80. J. Staudinger et al., “High efficien-cy CDMA RF power amplifier usingdynamic envelope tracking technique,”Int. Microwave Symp. Digest, vol. 2, pp.873-876, Boston, MA, June 13-15, 2000.
81. A. A. M. Saleh and D. C. Cox,“Improving the power-added efficiency ofFET amplifiers operating with varying-envelope signals,” IEEE Trans. MicrowaveTheory Tech., vol. 31, no. 1, pp. 51-55, Jan.1983.
82. B. D. Geller, F. T. Assal, R. K.Gupta, and P. K. Cline, “A technique forthe maintenance of FET power amplifierefficiency under backoff,” IEEE 1989MTT-S Digest, Long Beach, CA, pp. 949-952, June 1989.
83. G. Hanington, P.-F. Chen, P. M.Asbeck, and L. E. Larson, “High-efficiencypower amplifier using dynamic power-supply voltage for CDMA applications,”IEEE Trans. Micro-wave Theory Tech., vol.47, no. 8, pp. 1471-1476, Aug. 1999.
84. G. Hanington, A. Metzger, P.Asbeck, and H. Finlay, “Integrated dc-dcconverter using GaAs HBT technology,”Electronics Letters, vol. 35, no. 24, p.2110-2112, 1999.
85. D. R. Anderson and W. H. Cantrell,“High-efficiency inductor-coupled high-level modulator,” IMS ’01 Digest, Phoenix,AZ, May 2001.
86. H. Chireix, “High power outphas-ing modulation,” Proc. IRE, vol. 23, no. 11,pp. 1370-1392, Nov. 1935.
87. D. C. Cox and R. P. Leck, “A VHFimplementation of a LINC amplifier,”IEEE Trans. Commun., vol. COM-24, no.9, pp. 1018-1022, Sept. 1976.
88. R. Langridge, T. Thornton, P. M.Asbeck, and L. E. Larson, “A power re-usetechnique for improving efficiency of out-phasing microwave power amplifiers,”
IEEE Trans. Microwave Theory Tech., vol.47, no. 8, pp. 1467-1470 , Aug. 1999.
89. F. H. Raab, “Efficiency of outphas-ing power-amplifier systems,” IEEETrans. Commun., vol. COM-33, no. 10, pp.1094-1099, Oct. 1985.
90. B. Stengel and W. R. Eisenstat,“LINC power amplifier combiner efficien-cy optimization,” IEEE Trans. Veh.Technol., vol. 49, no. 1, pp. 229- 234, Jan.2000.
91. C. P. Conradi, R. H. Johnston, andJ. G. McRory, “Evaluation of a losslesscombiner in a LINC transmitter,” Proc.1999 IEEE Canadian Conf. Elec. andComp. Engr., Edmonton, Alberta, Canada,pp. 105-109, May 1999.
92. W. H. Doherty, “A new high effi-ciency power amplifier for modulatedwaves,” Proc. IRE, vol. 24, no. 9, pp. 1163-1182, Sept. 1936.
93. D. M. Upton and P. R. Maloney, “Anew circuit topology to realize high effi-ciency, high linearity, and high powermicrowave amplifiers,” Proc. Radio andWireless Conf. (RAWCON), ColoradoSprings, pp. 317-320, Aug. 9-12, 1998.
94. J. Schuss et al, “Linear amplifierfor high efficiency multi-carrier perfor-mance,” U.S. Patent 5,568,086, Oct. 1996.
95. J. Long, “Apparatus and methodfor amplifying a signal,” U.S. Patent5,886,575, March 1999.
96. F. H. Raab, “Efficiency of DohertyRF-power amplifier systems,” IEEETrans. Broadcasting, vol. BC-33, no. 3, pp.77-83, Sept. 1987.
97. M. Iwamoto et al., “An extendedDoherty amplifier with high efficiencyover a wide power range,” Int. MicrowaveSymp. Digest, Phoenix, AZ, pp. 931-934,May 2001.
98. B. E. Sigmon, “Multistage highefficiency amplifier,” U.S. Patent5,786,938, July 28, 1998.
Author InformationThe authors of this series of arti-
cles are: Frederick H. Raab (leadauthor), Green Mountain RadioResearch, e-mail: [email protected];Peter Asbeck, University of Californiaat San Diego; Steve Cripps, HywaveAssociates; Peter B. Kenington,Andrew Corporation; Zoya B. Popovic,University of Colorado; NickPothecary, Consultant; John F. Sevic,California Eastern Laboratories; andNathan O. Sokal, Design Automation.Readers desiring more informationshould contact the lead author.
Acronyms Used in Part 3
EER Envelope Elimination andRestoration
AM Amplitude Modulation
LINC Linear Amplification withNonlinear Components
38 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
RF and Microwave PowerAmplifier and TransmitterTechnologies — Part 4
By Frederick H. Raab, Peter Asbeck, Steve Cripps, Peter B. Kenington,Zoya B. Popovich, Nick Pothecary, John F. Sevic and Nathan O. Sokal
Linearization tech-niques are incorpo-rated into power
amplifiers and transmit-ters for the dual purposesof improving linearityand for allowing opera-tion with less back-offand therefore higher effi-ciency. This article pro-
vides a summary of the three main families oftechniques have been developed: Feedback,feedforward, and predistortion.
8a. FEEDBACKFeedback linearization can be applied
either directly around the RF amplifier (RFfeedback) or indirectly upon the modulation(envelope, phase, or I and Q components).
RF FeedbackThe basis of this technique is similar to its
audio-frequency counterpart. A portion of theRF-output signal from the amplifier is fedback to, and subtracted from, the RF-inputsignal without detection or down- conversion.Considerable care must be taken when usingfeedback at RF as the delays involved must besmall to ensure stability. In addition, the lossof gain at RF is generally a more significantsacrifice than it is at audio frequencies. Forthese reasons, the use of RF feedback in dis-crete circuits is usually restricted to HF andlower VHF frequencies [99]. It can be appliedwithin MMIC devices, however, well into themicrowave region.
In an active RF feedback system, the volt-age divider of a conventional passive-feedbacksystem is replaced by an active (amplifier)
stage. The gain in the feedback path reducesthe power dissipated in the feedback compo-nents. While such systems demonstrate IMDreduction [105], they tend to work best at aspecific signal level.
Envelope FeedbackThe problem of delay in RF feedback is
alleviated to a large extent by utilizing thesignal envelope as the feedback parameter.This approach takes care of in-band distortionproducts associated with amplitude nonlin-earity. Harmonic distortion products, whichare corrected by RF feedback, are generallynot an issue as they can easily be removed byfiltering in most applications. Envelope feed-back is therefore a popular and simple tech-nique.
Envelope feedback can be applied to eithera complete transmitter (Figure 41) or a singlepower amplifier (Figure 42). The principles ofoperation are similar and both are describedin detail in [100]. The RF input signal is sam-pled by a coupler and the envelope of the inputsample is detected. The resulting envelope isthen fed to one input of a differential amplifi-er, which subtracts it from a similarly
Linearization methods arethe focus of Part 4 of our
series on power amplifiers,which describes the basic
architecture and perfor-mance capabilities of feed-
back, feedforward andpredistortion techniques Fig 41 · Envelope feedback applied to a
obtained sample of the RF output. The difference signal,representing the error between the input and outputenvelopes, is used to drive a modulator in the main RFpath. This modulator modifies the envelope of the RF sig-nal which drives the RF PA. The envelope of the resultingoutput signal is therefore linearized to a degree deter-mined by the loop gain of the feedback process. Examplesof this type of system are reported in [101] and [102].
The degree of linearity improvement that can beobtained when using this technique depends upon the rel-ative levels of the AM-AM and AM-PM conversion in theamplifier. For a VHF BJT amplifier, AM-AM distortion isdominant and two-tone IMD is typically reduced by 10dB. Since AM-PM distortion is not corrected by envelopefeedback, no linearity improvement is observed if phasedistortion is the dominant form of nonlinearity. This isoften the case in, for example, class-C and LDMOS PAs.The use of envelope feedback is therefore generallyrestricted to relatively linear class-A or AB amplifiers.
Polar-Loop FeedbackThe polar-loop technique overcomes the fundamental
inability of envelope feedback to correct for AM-PM dis-tortion effects [103]. Essentially, a phase-locked loop isadded to the envelope feedback system as shown inFigure 43. For a narrowband VHF PA, the improvementin two-tone IMD is typically around 30 dB.
The envelope- and phase-feedback functions operateessentially independently. In this case, envelope detectionoccurs at the intermediate frequency (IF), as the inputsignal is assumed to be a modulated carrier at IF.Likewise, phase detection takes place at the IF, with lim-iting being used to minimize the effects of signal ampli-tude upon the detected phase. Alternatively, it is possibleto supply the envelope and phase modulating signals sep-arately at baseband and to undertake the comparisonsthere.
The key disadvantage of polar feedback lies in the gen-
erally different bandwidths required for the amplitudeand phase feedback paths. Thus, differing levels ofimprovement of the AM-AM and AM-PM characteristicsusually result, and this often leads to a poorer overall per-formance than that achievable from an equivalentCartesian-loop transmitter. A good example of the differ-ence occurs with a standard two-tone test, which causesthe phase-feedback path to cope with a discontinuity atthe envelope minima. In general, the phase bandwidthmust be five to ten times the envelope bandwidth, whichlimits available loop gain for a given delay.
Cartesian FeedbackThe Cartesian-feedback technique overcomes the
problems associated with the wide bandwidth of the sig-nal phase by applying modulation feedback in I and Q(Cartesian) components [104]. Since the I and Q compo-nents are the natural outputs of a modern DSP, theCartesian loop is widely used in PMR and SMR systems.
Figure 42 · Envelope feedback applied to an RF poweramplifier.
Figure 43 · Block diagram of a polar-loop transmitter.
The basic Cartesian loop (Figure 44) consists of twoidentical feedback processes operating independently onthe I and Q channels. The inputs are applied to differen-tial integrators (in the case of a first-order loop) with theresulting difference (error) signals being modulated ontoI and Q subcarriers and up-converted to drive the PA. Asample of the output from the PA is attenuated andquadrature-down-converted (synchronously with the up-conversion process). The resulting quadrature feedbacksignals then form the second inputs to the input differen-tial integrators, completing the two feedback loops. Thephase shifter shown in the up-converter local-oscillatorpath is used to align the phases of the up- and down-con-version processes, thereby ensuring that a negative feed-back system is created and that the phase margin of thesystem is optimized.
The effects of applying Cartesian feedback to a highlynonlinear (class-C) PA amplifying an IS-136 (DAMPS)signal are shown in Figure 45. The first ACPR isimproved by 35 dB and the signal is produced withinspecifications with an efficiency of 60 percent [100].
8b. FEEDFORWARDThe very wide bandwidths (10 to 100 MHz) required in
multicarrier applications can render feedback and DSPimpractical. In such cases, the feedforward technique canbe used to achieve ultra-linear operation. In its basic con-figuration, feedforward typically gives improvements indistortion ranging from 20 to 40 dB.
Operation In its basic form (Figure 46), a feedforward amplifier
consists of two amplifiers (the main and error amplifiers),directional couplers, delay lines and loop control networks[110]. The directional couplers are used for power split-
ting/combining, and the delay lines ensure operation overa wide bandwidth. Loop-control networks, which consistof amplitude- and phase-shifting networks, maintain sig-nal and distortion cancellation within the various feed-forward loops.
The input signal is first split into two paths, with onepath going to the high-power main amplifier while theother signal path goes to a delay element. The output sig-nal from the main amplifier contains both the desired sig-nal and distortion. This signal is sampled and scaledusing attenuators before being combined with the delayedportion of the input signal, which is regarded as distor-tion-free. The resulting “error signal” ideally containsonly the distortion components in the output of the mainamplifier. The error signal is then amplified by the low-power, high-linearity error amplifier, and then combinedwith a delayed version of the main amplifier output. Thissecond combination ideally cancels the distortion compo-nents in the main-amplifier output while leaving thedesired signal unaltered.
In practice, there is always some residual desired sig-nal passing through the error amplifier. This is in gener-al not a problem unless the additional power is sufficientin magnitude to degrade the linearity of the error ampli-fier and hence the linearity of the feedforward transmit-ter.
Signal CancellationSuccessful isolation of an error signal and the removal
of distortion components depend upon precise signal can-cellation over a band of frequencies. In practice, cancella-tion is achieved by the vector addition of signal voltages.The allowable amplitude and phase mismatches for dif-ferent cancellation levels are shown in Figure 47. Formanufactured equipment, realistic values of distortion
Figure 45 · Linearization of a class-C PA byCartesian feedback (courtesy WSI).
Figure 46 · Block diagram of a feed-forward transmitter in itsbasic form.
44 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
cancellation are around 25 to 30. The limiting factor isnearly always the bandwidth over which a given accura-cy can be obtained.
EfficiencyThe outputs of the main and error amplifiers are typ-
ically combined in a directional coupler that both isolatesthe PAs from each other and provides resistive inputimpedances. For a typical 10 dB coupling ratio, 90 percentof the power from the main PA reaches the output. For thesame coupling ratio, only 10 percent of the power from theerror amplifier reaches the load, thus the error amplifiermust produce ten times the power of the distortion in themain amplifier. The peak-to-average ratio of the error sig-nal is often much higher than that of the desired signal,making amplification of the error signal inherently muchless efficient than that of the main signal. As a result, thepower consumed by the error amplifier can be a signifi-cant fraction (e.g., one third) of that of the main amplifi-er. In addition, it may be necessary to operate one or bothamplifiers well into back-off to improve linearity. Theoverall average efficiency of a feedforward transmittermay therefore be only 10 to 15 percent for typical multi-carrier signals.
Automatic Loop ControlSince feedforward is inherently an open-loop process,
changes in device characteristics over time, temperature,voltage and signal level degrade the amplitude and phasematching and therefore increase distortion in the trans-mitter output. An automatic control scheme continuouslyadjusts the gain and phase to achieve the best signal can-cellation and output linearity. The first step is to use FFTtechniques, direct power measurement, or pilot signals todetermine how well the loop is balanced. Both digital andanalog techniques can be used for loop control and adjust-ment. Signal processing can be used to reduce the peaksin multi-carrier signals and to keep distortion productsout of the nearby receiving band [111].
PerformanceAn example of the use of feedforward to improve lin-
earity is shown in Figure 48. The signal consists of a mixof TDMA and CDMA carriers. The power amplifiers arebased upon LDMOS transistors and have two-tone IMDlevels in the range –30 to –35 dBc at nominal outputpower. The addition of feedforward reduces the level ofdistortion by approximately 30 dB to meet the requiredlevels of better than –60 dBc. The average efficiency istypically about 10 percent.
8c. PREDISTORTIONThe basic concept of a predistortion system (Figure 49)
involves the insertion of a nonlinear element prior to the
Figure 47 · Gain/phase matching requirements.Figure 48 · Feedforward performance with mixed-mode modulation (TDMA and CDMA signals).
RF PA such that the combined transfer characteristic ofboth is linear (Figure 50). Predistortion can be accom-plished at either RF or baseband.
RF PredistortionThe block diagram of a simple RF predistorter is
shown in Figure 51. A compressive characteristic, createdby the nonlinearity in the lower path (e.g., a diode) is sub-tracted from a linear characteristic (the upper path) togenerate an expansive characteristic. The output of thelinear path (typically just a time delay) is given by:
vl(vin) = a1vin (1)
and that of the compressive path is given by
vc(vin) = a2vin – bv3in (2)
Subtracting the above equations gives
vpd(vin) = (a2 – a2)vin – bv3in (3)
This is now an expansive characteristic with a lineargain of a1 – a2, and may be used to predistort a compres-sive amplifier characteristic (cubic in this example) byappropriate choice of a1, a2 and b.
An example of the results from using a simple diode-based RF predistorter with a 120-W LDMOS PA amplify-ing an IS-95 CDMA signal is shown Figure 52. Whenapplied to π/4-DQPSK modulation in a satellite applica-tion, the same predistorter roughly halves the EVM,improves the efficiency from 22 to 29 percent, and doublesthe available output power.
Predistortion bandwidths tend to be limited by similarfactors to that of feedforward, namely gain and phaseflatness of the predistorter itself and of the RF PA. Inaddition, memory effects in the PA and the predistorterlimit the degree cancellation, and these tend to becomepoorer with increasing bandwidth.
Better performance can be achieved with more com-plex forms of RF predistortion such as Adaptive
Parametric Linearization (APL®), which is capable ofmulti-order correction [106]. Most RF-predistortion tech-niques are capable of broadband operation with practicaloperational bandwidths similar to, or greater than, thoseof feedforward.
Digital Predistortion Digital predistortion techniques exploit the consider-
able processing power now available from DSP devices,which allows them both to form and to update therequired predistortion characteristic. They can operatewith analog-baseband, digital-baseband, analog-IF, digi-tal-IF, or analog-RF input signals. Digital-baseband anddigital-IF processing are most common.
The two most common types of digital predistorter aretermed mapping predistorters [107] and constant-gain
Figure 51 · An RF predistorter.
Figure 52 · Linearization by diode-based RF predistorter(courtesy WSI).
Figure 53 · Mapping predistorter.
48 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
predistorters [108]. A mapping predistorter utilizes twolook-up tables, each of which is a function of two variables(IIN and QIN), as shown in Figure 53. This type of predis-torter is capable of excellent performance. However, itrequires a significant storage and/or processing overheadfor the look-up tables and their updating mechanism, andhas a low speed of convergence. The low convergencespeed results from the need to address all points in theI/Q complex plane before convergence can be completed.
A constant-gain predistorter (Figure 54) requires onlya single-dimensional look-up table, indexed by the signalenvelope. It is therefore a much simpler implementationand requires significantly less memory for a given level ofperformance and adaptation time. It uses the look-uptable to force the predistorter and associated PA to exhib-it a constant gain and phase at all envelope levels. The
overall transfer characteristic is then linear:
GPD(IIN(t),QIN(t))×GPA(IPD(t),QPD(t)) = k (4)
An example of the improvement in the amplitude-transfer characteristic by an RF-input/output digital pre-distorter [109] is shown in Figure 55. The plot is basedupon real-time using samples from a GSM-EDGE signal.Both the gain expansion and compression are improvedby the linearizer. EVM is reduced from around 4.5 to 0.7percent. The ACPR for IS-136 DAMPS modulation (π/4-DQPSK) is reduced by nearly 20 dB (Figure 56). Whengenerating mask-compliant EDGE modulation at full out-put power (850-900 MHz), the linearized PA has an effi-ciency of over 30 percent.
An example of linearization of a PA with two 3G W-
Figure 54 · Constant-gain predistorter.
Figure 55 · Linearization of the amplitude transfer char-acteristic using an RF input/output digital predistorter(courtesy WSI).
Figure 56 · Linearization of DAMPS PA by RF input/out-put predistorter (courtesy WSI).
Figure 57 · Linearization of 3G W-CDMA PA signal bydigitial baseband input predistorter (courtesy WSI).
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
00 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Input envelope2
Ou
tpu
t e
nve
lop
e2
November 2003 49
CDMA signals by a digital baseband-input predistorter isshown in Figure 57. The linearized amplifier meets therequired spectral mask with a comfortable margin at allfrequency offsets. The noise floor is set by the degree ofclipping employed on the waveform, which limits theACPR improvement obtained. It clearly demonstrates,however, that digital predistortion can be used in broad-band as well as narrowband applications. Figure 58shows a 3G transmitter that uses digital predistortion.
References99. A. F. Mitchell, “A 135 MHz feedback amplifier,” IEE
Colloq. Broadband High Frequency Amplifiers: Practiceand Theory, pp. 2/1-2/6, London, Nov. 22, 1979
100. P. B. Kenington, High Linearity RF AmplifierDesign, Norwood, MA: Artech, 2000.
101. W. B. Bruene, “Distortion reducing means for sin-gle-sidedband transmitters,” Proc. IRE, vol. 44, no. 12, pp.1760-1765, Dec. 1956.
102. T. Arthanayake and H. B. Wood, “Linear amplifi-cation using envelope feedback,” Electronics Letters, vol.7, no. 7, pp. 145-146, April 8, 1971.
103. V. Petrovic and W. Gosling, “Polar-loop transmit-ter,” Electronics Letters, vol. 15, no. 10, pp. 286-287, May10, 1979.
104. V. Petrovic, “Reduction of spurious emission fromradio transmitters by means of modulation feedback,”Proc. IEE Conf. No. 224 on Radio Spectrum ConservationTechniques, UK., Sept. 6-8 1983.
105. E. Ballesteros, F. Perez, and J. Peres, “Analysisand design of microwave linearized amplifiers usingactive feedback,” IEEE Trans. Microwave Theory Tech.,vol. 36, no. 3, pp. 499-504, March 1988.
106. P. B. Kenington, “Achieving high-efficiency inmulti-carrier base-station power amplifiers,” MicrowaveEngr. Europe, pp. 83-90. Sept. 1999.
107. Y. Nagata, “Linear amplification techniques fordigital mobile communications,” Proc. IEEE Veh. Tech.Conf. (VTC ’89), San Fransisco, pp. 159-164, May 1-3,1989.
108. J. K. Cavers, “Amplifier linearisation using a dig-ital predistorter with fast adaptation and low memoryrequirements,” IEEE Trans. Veh. Tech., vol. 39, no. 4, pp.374-382, Nov. 1990.
109. P. B. Kenington, M. Cope, R. M. Bennett, and J.Bishop, “GSM-EDGE high power amplifier utilising digi-tal linearisation,” IMS’01 Digest, Phoenix, AZ, May 20-25,2001.
110. N. Pothecary, Feedforward Linear PowerAmplifiers, Norwood, MA: Artech, 1999.
111. J. Tellado, Multicarrier Modulation with LowPAR, Boston: Kluwer, 2000.
Author InformationThe authors of this series of articles are: Frederick H.
Raab (lead author), Green Mountain Radio Research, e-mail: [email protected]; Peter Asbeck, University ofCalifornia at San Diego; Steve Cripps, Hywave Associates;Peter B. Kenington, Andrew Corporation; Zoya B. Popovic,University of Colorado; Nick Pothecary, Consultant; JohnF. Sevic, California Eastern Laboratories; and Nathan O.Sokal, Design Automation. Readers desiring more infor-mation should contact the lead author.
Figure 58 · A multi-carrier S-band transmitter with digi-tal predistorter (courtesy WSI).
Acronyms Used in Part 4
ACPR Adjacent Channel Power RatioAPL Adaptive Parametric LinearizationBER Bit Error RateDAMPS Digital American Mobile Phone SystemEDGE Enhanced Data for GSM EvolutionEVM Error Vector MagnitudeIF Intermediate FrequencyLDMOS Laterally Diffused Metal Oxide SemiconductorPA Power AmplifierPDF Probability-Density FunctionPMR Private Mobile RadioSMR Specialized Mobile RadioW-CDMA Wideband Code-Division Multiple Access
46 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
RF and Microwave PowerAmplifier and TransmitterTechnologies — Part 5
By Frederick H. Raab, Peter Asbeck, Steve Cripps, Peter B. Kenington,Zoya B. Popovich, Nick Pothecary, John F. Sevic and Nathan O. Sokal
The ever-increasingdemands for more band-width, coupled withrequirements for bothhigh linearity and highefficiency create ever-increasing challenges inthe design of poweramplifiers and transmit-ters. A single W-CDMA
signal, for example, taxes the capabilities of aKahn-technique transmitter with a conven-tional class-S modulator. More acute are theproblems in base-station and satellite trans-mitters, where multiple carriers must beamplified simultaneously, resulting in peak-to-average ratios of 10 to 13 dB and band-widths of 30 to 100 MHz.
A number of the previously discussed tech-niques can be applied to this problem, includingthe Kahn EER with class-G modulator or split-band modulator, Chireix outphasing, andDoherty. This section presents some emergingtechnologies that may be applied to wideband,high efficiency amplification in the near future.
RF Pulse-Width ModulationVariation of the duty ratio (pulse width) of
a class-D RF PA [112] produces an amplitude-modulated carrier (Figure 59). The outputenvelope is proportional to the sine of thepulse width, hence the pulse width is varied inproportion to the inverse sine of the desiredenvelope. This can be accomplished in DSP, orby comparison of the desired envelope to afull-wave rectified sinusoid. The pulse timing
conveys signal phase information as in theKahn and other techniques.
Radio-frequency pulse-width modulation(RF PWM) eliminates the series-pass lossesassociated with the class-S modulator in aKahn-technique transmitter. More important-ly, the spurious products associated withPWM are located in the vicinity of the har-monics of the carrier [113] and therefore easi-ly removed. Consequently, RF PWM canaccommodate a significant RF bandwidthwith only a simple, low-loss output filter.
Ideally, the efficiency is 100 percent. Inpractice, switching losses are increased overthose in a class-D PA with a 50:50 duty ratiobecause drain current is nonzero duringswitching.
Emerging techniques areexamined in this final
installment of our series onpower amplifier technolo-
Previous applications of RF PWM have been limited toLF and MF transmitters (e.g., GWEN [114]). However, therecent development of class-D PAs for UHF andmicrowave frequencies (Figure 60) offers some interestingpossibilities.
Delta-Sigma ModulationDelta-sigma modulation is an alternative technique
for directly modulating the carrier produced by a class-DRF PA (Figure 61) [PA8],[PA9]. In contrast to the basical-ly analog operation of RF PWM, delta-sigma modulationdrives the class-D PA at a fixed clock rate (hence fixedpulse width) that is generally higher than the carrier fre-quency (Figure 62). The polarity of the drive is toggled asnecessary to create the desired output envelope from the
average of the cycles in the PA. Phase is again conveyedin pulse timing.
The delta-sigma modulator employs an algorithmsuch as that shown in Figure 63. The signal is digitized bya quantizer (typically a single-bit comparator) whose out-put is subtracted from the input signal through a digitalfeedback loop, which acts as a band-pass filter. Basically,the output signal in the pass band is forced to track thedesired input signal. The quantizing noise (associatedwith the averaging process necessary to obtain thedesired instantaneous output amplitude) is forced outsideof the pass band.
The degree of suppression of the quantization noisedepends on the oversampling ratio; i.e., the ratio of thedigital clock frequency to the RF bandwidth and is rela-tively independent of the RF center frequency. An exam-ple of the resultant spectrum for a single 900-MHz carri-er and 3.6-GHz clock is shown in Figure 64. The quanti-zation noise is reduced over a bandwidth of 50 MHz,which is sufficient for the entire cellular band. Out-of-band noise increases gradually and must be removed by aband-pass filter with sufficiently steep skirts.
As with RF PWM, the efficiency of a practical delta-sigma modulated class-D PA is reduced by switching loss-es associated with nonzero current at the times of switch-ing. The narrow-band output filter may also introducesignificant loss.
Carrier Pulse-Width ModulationCarrier pulse-width modulation was first used in a
UHF rescue radio at Cincinnati Electronics in the early1970s. Basically, pulse-width modulation as in a class-Smodulator gates the RF drive (hence RF drain current) onand off in bursts, as shown in Figure 65. The width of eachburst is proportional to the instantaneous envelope of the
Figure 60 · Current-switching PA for 1 GHz (courtesyUCSD).
Figure 61 · Prototype class-D PA for delta-sigma mod-ulation (courtesy UCSD).
Figure 62 · Delta-sigma modulation.
January 2004 49
desired output. The amplitude-modulated output signal isrecovered by a band-pass filter that removes the side-bands associated with the PWM switching frequency. ThePWM signal can be generated by a comparator as in aclass-S modulator or by delta-sigma techniques.
As with RF PWM and delta-sigma modulation, theseries-pass losses and bandwidth limitations of the high-level modulator are eliminated. The switching frequencyin carrier PWM is not limited by capabilities of power-switching devices and can therefore easily be tens ofMHz, allowing large RF bandwidths. A second advantageis that carrier PWM can be applied to almost any type ofRF PA. A disadvantage is that a narrow-band output fil-ter with steep skirts is required to remove the switching-frequency sidebands, and such filters tend to have lossesof 1 to 2 dB at microwave frequencies. Nonetheless, thelosses in the filter may be more than offset by theimprovement in efficiency for signals with high peak-to-average ratios.
Power RecoveryA number of RF processes result in significant RF
power dissipated in “dump” resistors. Examples includepower reflected from a mismatched load and dumped by acirculator and the difference between two inputs of hybridcombiner dumped to the balancing resistor. The notion ofrecovering and reusing wasted RF power was originallyapplied to the harmonics (18 percent of the output power)of an untuned LF class-D PA [117].
More recently, power recovery has been applied to out-phasing PAs with hybrid combiners [118, 119]. Theinstantaneous efficiency of such a system depends uponboth the efficiency of the PA and that of the recovery sys-tem. Since the two PAs operate at full power regardless ofthe system output, inefficiency in the PA has a significantimpact upon the system efficiency at the lower outputs.Nonetheless, a significant improvement over convention-al hybrid-coupled outphasing is possible, and the PAs arepresented with resistive loads that allow them to operateoptimally. Typically, 50 percent of the dumped power canbe recovered.
The power-recovery technology can also be used toimplement miniature DC-DC converters. Basically, ahigh-efficiency RF-power amplifier (e.g., class-E) convertsDC to RF and a high-efficiency rectifier circuit convertsthe RF to DC at the desired voltage. Implementation atmicrowave frequencies reduces the size of the tuning andfiltering components, resulting in a very small physicalsize and high power density. In a prototype that operatesat C band [120], the class-E PA uses a single MESFET toproduce 120 mW with a PAE of 86 percent. The diode rec-tifier consists of a directional coupler with two Schottky
Figure 65 · Carrier pulse-width modulation.
Figure 63 · Delta-sigma modulator.
Figure 64 · Spectrum of delta-sigma modulation.
–30
–40
–50
–60
–70
–80
–90
–100
Po
we
r Sp
ec
tra
l De
nsit
y (d
Bm/H
z)
600 800 1000 1200
Frequency (MHz)
50 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
diodes connected at the coupled and through ports andhas a 98-percent conversion efficiency and an overall effi-ciency (including mismatch loss) of 83 percent. For a typ-ical DC output of 3 V, the DC-DC conversion efficiency is64 percent.
Switched PAs with Transmission-Line CombinersRF-power amplifiers cannot simply be connected in
series or parallel and switched on and off to make a trans-mitter module that adapts to variable peak envelopepower. Attempting to do so generally produces either lit-tle effect or erratic variations in load impedance, some-times leading to unstable operation and destruction of thetransistors. Systems of microwave PAs that are toggledon and off are therefore connected through networks ofquarter-wavelength transmission lines. The Dohertytransmitter (discussed in part 4 of this series) is a classicexample of this sort of technique.
An alternative topology (Figure 66) uses shortingswitches and quarter-wavelength lines to to decouple off-state PAs [121, 122]. The inactive PA is powered-down (byswitching off its supply voltage), after which its output isshorted to ground. The quarter-wavelength line producesan open circuit at the opposite end where the outputsfrom multiple PAs are connected together to the load.This technique may be more easy to implement (especial-ly for multiple PAs) than Doherty because a short is morereadily realized than an open.
If PA #1 is the only PA active, its load is simply R0. Ifboth #1 and #2 are active, the combination produces aneffective load impedance of 2Ro at the load ends of thelines. Inversion of this impedance through the linesplaces loads of R0/2 on the RF PAs. Consequently, thepeak power output for two active PAs is four times thatwith a single PA. As in discrete envelope tracking, the RFPAs operate as linear amplifiers. The number of PAs that
are active is the minimum needed to produce the currentoutput power. The peak power is thus kept relatively closeto the saturated output, eliminating most of the effects ofoperating in back-off. The efficiency can therefore reachPEP efficiency at a number of different output levels, asshown in Figure 67.
The advantage of this technique is the ease in designassociated with relying on short circuits rather than opencircuits to isolate the off-state PAs. A possible disadvan-tage is operating individual PAs from multiple loadimpedances without retuning and a limited number ofpower steps available (e.g., 9/9, 4/9, 1/9 for a three-PA sys-tem).
Electronic TuningThe performance of virtually all power amplifiers is
degraded by load- impedance mismatch. Mismatchedloads not only reduce efficiency, but also create higherstresses on the transistors. Because high-efficiency PAsgenerally require a specific set of harmonic impedances,their use is often restricted to narrow-band applicationswith well-defined loads.
Electronic tuning allows frequency agility, matching ofunknown and variable loads, and amplitude modulation.Components for electronic tuning include pin-diodeswitches, MEMS switches, MEMS capacitors, semicon-ductor capacitors, ceramic capacitors (e.g., BST), and bias-controlled inductors. To date, electronic tuning has beenapplied mainly to small-signal circuits such as voltage-controlled oscillators. Recently demonstrated, however,are two electronically tuned power amplifiers. One oper-ates in class E, produces 20 W with an efficiency of 60 to70 percent, and can be tuned from 19 to 32 MHz (1.7:1range) through the use of voltage-variable capacitors
Figure 66 · Switched PAs with quarter-wavelengthtransmission line combiner.
Figure 67 · Instantaneous efficiency of switched PAs.
52 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
[123, 124]. The second (Figure 68) operates in class D, pro-duces 100 W with an efficiency of 60 to 70 percent, andcan be tuned from 5 to 21 MHz (4.25:1 range) through theuse of electronically tunable inductors and capacitors[125].
Load ModulationThe output of a power amplifier can be controlled by
varying the drive, gate bias, DC supply voltage, or loadimpedance. “Load modulation” uses an electronicallytuned output filter (Figure 69) to vary load impedanceand thereby the instantaneous amplitude of the outputsignal. The modulation bandwidth can be quite wide, as itis limited only by the bias feeds to the tuning components.
A key aspect of load modulation is a diligent choice ofthe impedance locus so that it provides both good dynam-ic range and good efficiency. For ideal saturated PAs ofclasses A, B, C, and F, the optimum locus is the pure resis-tance line on the Smith chart that runs from the nominalload to an infinite load. For ideal class-E PAs with seriesinductance and shunt susceptance for optimum operationwith the nominal load, the optimum locus is the unity-efficiency line running from the nominal load upward andrightward at an angle of 65° [126]. For real PAs, the opti-
mum locus is found by examination of load-pull contours.The simple T filter has a single electronically variable
element, but provides an approximately optimum locusfor class E over the top 12 dB of the dynamic range. Theexperimental 20-W, 30-MHz [124, 126] shown in Figure70 achieves a 41-dB range of amplitude variation. Themeasured instantaneous-efficiency curve (Figure 71) cor-responds to a factor of 2.1 improvement in the averageefficiency for a Rayleigh-envelope signal with a 10-dBpeak-to-average ratio.
A load-modulated PA for communications follows theelectronically tuned filter with a passive filter to removethe harmonics associated with the nonlinear elements.Predistortion compensates for the incidental phase mod-ulation inherent in dynamic tuning of the filter. Variationof the drive level can be used to conserve drive power andto extend the dynamic range.
Figure 70 · Load-modulated class-E PA (courtesyGMRR).
Figure 71 · Instantaneous efficiency of load modula-tion compared to class-B linear amplification.
Figure 68 · Electronically tunable class-D PA (courtesyGMRR).
Figure 69 · Load modulation by electronic tuning.
54 High Frequency Electronics
High Frequency Design
RF POWER AMPLIFIERS
References112. Figures 8 and 9 in Part 2 of
tis series, High FrequencyElectronics, July 2003.
113. F. H. Raab, “Radio frequencypulsewidth modulation,” IEEE Trans.Commun., vol. COM-21, no. 8, pp.958-966, Aug. 1973.
114. F. G. Tinta, “Direct singlesideband modulation of transmitteroutput switcher stages,” Proc. RFExpo East ’86, Boston, MA, pp. 313-398, Nov. 10-12, 1986.
115. A. Jayaraman, P. F. Chen, G.Hanington, L. Larson, and P. Asbeck,“Linear high-efficiency microwavepower amplifiers using bandpassdelta-sigma modulators,” IEEEMicrowave Guided Wave Lett., vol. 8,no. 3, pp. 121-123, March 1998.
116. J. Keyzer et al., “Generationof RF pulsewidth modulatedmicrowave signals using delta-sigmamodulation,” 2002 Int. MicrowaveSymp. Digest, vol. 1, pp. 397-400,Seattle, WA, June 2-7, 2002.
117. J. D. Rogers and J. J. Wormser,“Solid-state high-power low frequencytelemetry transmitters,” Proc. NEC,vol. 22, pp. 171-176, Oct. 1966.
118. R. E. Stengel and S. A. Olson,“Method and apparatus for efficientsignal power amplification,” U.S.Patent 5,892,395, Apr. 6, 1999.
119. R. Langridge, T. Thornton, P.M. Asbeck, and L. E. Larson, “A powerre-use technique for improving effi-ciency of outphasing microwavepower amplifiers,” IEEE Trans.Microwave Theory Tech., vol. 47, no. 8,pp. 1467-1470 , Aug. 1999.
120. S. Djukic, D. Maksimovic, andZ. Popovic, “A planar 4.5-GHz dc-dcpower converter,” IEEE Trans.Microwave Theory Tech., vol. 47, no. 8,pp. 1457-1460, Aug. 1999.
121. A. Shirvani, D. K. Su, and B.A. Wooley, “A CMOS RF power ampli-fier with parallel amplification forefficient power control,” IEEE J.Solid- State Circuits, vol. 37, no. 6,pp. 684-693, June 2002.
122. C. Y. Hang, Y. Wang, and T.Itoh, “A new amplifier power combin-
ing scheme with optimum efficiencyunder variable outputs,” 2002 Int.Microwave Symp. Digest, vol. 2, pp.913-916, Seattle, WA, June 2-7, 2002.
123. F. H. Raab, “Electronicallytunable class-E power amplifier,” Int.Microwave Symp. Digest, Phoenix,AZ, pp. 1513-1516, May 20-25, 2001.
124. F. H. Raab, “Electronicallytuned power amplifier,” Patent pend-ing.
125. F. H. Raab and D. Ruppe,“Frequency-agile class-D poweramplifier,” Ninth Int. Conf. on HFRadio Systems and Techniques, pp.81-85, University of Bath, UK, June23-26, 2003.
126. F. H. Raab, “High-efficiencylinear amplification by dynamic loadmodulation,” Int. Microwave Symp.Digest, vol. 3, pp. 1717-1720,Philadelphia, PA, June 8-13, 2003.
CorrectionIn Part 4 of this series (Novermber2003 issue), Figures 45, 52, 55, 56, 57and 58 should have been credited as“Courtesy Andrew Corporation”instead of “Courtesy WSI.”
Author InformationThe authors of this series of arti-
cles are: Frederick H. Raab (leadauthor), Green Mountain RadioResearch, e-mail: [email protected];Peter Asbeck, University of Californiaat San Diego; Steve Cripps, HywaveAssociates; Peter B. Kenington,Andrew Corporation; Zoya B. Popovic,University of Colorado; NickPothecary, Consultant; John F. Sevic,California Eastern Laboratories; andNathan O. Sokal, Design Automation.Readers desiring more informationshould contact the lead author.
