High Frequency Design From December 2008 High Frequency Electronics

Copyright © 2008 Summit Technical Media, LLC

DOHERTY AMPLIFIER

A High Efficiency Doherty

Amplifier with Digital
Predistortion for WiMAX
By Simon Wood and Ray Pengelly, Cree, Inc.,
and Jim Crescenzi, Central Coast Microwave Design, LLC

T
he Doherty ampli-
This article describes a fier architecture is
WiMAX power amplifier , a well known tech-
which achieves high nique offering the poten-
performance using the tial to improve transmit-
latest device technologies ter efficiency especially
and design techniques for signal protocols that
exhibit high peak-to-
average power ratios. Although the Doherty
approach has significant efficiency advan-
tages, it generally must be augmented with
some form of correction or linearity enhance- Figure 1 · A basic 2-way Doherty amplifier
ment in the full transmitter design. Wireless configuration.
infrastructure applications have demanding
linearity and spectral mask specifications. The
latest WiMAX standards present a particular emissions mask (SEM) at close offsets (1.5
challenge with their combination of very high MHz) be at least –45 dB. A typical amplifier
peak-to-average power ratios, 10 MHz or (without correction) will exceed this level by 15
wider channel bandwidths and high linearity to 20 dB, resulting in a significant correction
standards. This article demonstrates a 2.5-2.7 demand for the digital predistortion circuitry.
GHz Doherty amplifier that achieves, when The technologies applicable to this chal-
augmented with digital predistortion, 8 watts lenge have improved in three areas:
of WiMAX average output power at greater
than 47% efficiency, while satisfying demand- • Improved GaN HEMT transistors with
ing spectral mask specifications. reduced capacitances and trapping effects
The appeal of the Doherty amplifier config- • Improved digital predistortion subsys-
uration is that it involves familiar core amplifi- tems and software
er designs connected in a manner that main- • Doherty circuit design refinements that
tains high efficiency over an extended input reduce memory effects, increase gain and
signal range. Doherty development has been increase bandwidth.
energized by the latest generations of transis-
tors and is well represented in recent literature This article focuses on the optimization of
[1-5]. Applying the Doherty approach to mod- Doherty amplifier circuitry for WiMAX appli-
ern WiMAX signals can present unique chal- cations.
lenges. In particular, WiMAX signals combine
two challenging features: wide video bandwidth Amplifier Design Approach
of 10 MHz or greater, and large peak-to-average Doherty amplifiers can be configured with
ratios of 10 dB or greater. Additionally, the lat- 2-, 3- or N-way combinations [8, 9, 10]. The
est WiMAX standards require that the spectral most commonly used Doherty architecture

18 High Frequency Electronics

 High Frequency Design
DOHERTY AMPLIFIER

involves two amplifiers, as shown in

Figure 1. Adding additional branches
can extend the power range over
which high efficiency is maintained.
However, the 2-amplifier approach is
often preferred due to cost considera-
tions (i.e., using fewer devices).
Fortunately, there are multiple vari-
ables besides the number of branches
that can be adjusted to optimize per- Figure 2 · Schematic of an unequal Figure 3 · Simulated gain and effi-
formance. power divider for Doherty applica- ciency vs. CW output power for two
The Doherty block diagram (Fig. tion. input couplers. Dark trace is for 3 dB
1) is notable for its elegance and sim- (equal) coupling. Light trace is for
plicity. What may not be apparent is approximately 1.8/4.7 dB coupling
that details of the design can result amplifier efficiency over a power (more power to carrier amplifier
in large differences in performance. range from 10 dB backoff to peak versus peaking amplifier).
Operation is strongly influenced by (saturated) output level.
the coupling factor of the input Series matching elements of each
divider/coupler and by the biasing of stage were modified to reduce electri- because of layout constraints. This
the carrier and peaking amplifier cal length from the transistor to the divider maintains a reasonable level
stages. The “turn-on” of the peaking combining node, and harmonic termi- of isolation, while being more com-
amplifier is dependent on both input nations were adjusted to optimize the pact than if it were extended to
power level and gate bias voltage, designs under Doherty operation. accommodate a second resistor. The
which in turn sets the low power effi- It has been our experience that delay line is added to provide the
ciency and peak power capability of circuit modifications required after desired quadrature relationship of
the configuration. The peaking insertion into the Doherty configura- the two outputs. The coupling is
amplifier allows the Doherty amplifi- tion have been minimal for the input determined by the relative line
er to respond to the high input levels matching circuits, but substantial for widths (impedances) of the two
of short duration by amplifying the the output circuits. Inserting the pre- branches. The divider block could be
signal peaks and dynamically chang- ferred stage designs as carrier and further compacted by substituting a
ing the loading on the main amplifier. peaking stages (optimized for stand- custom coupled-line design with over-
The specific class of operation alone operation in a 50-ohm environ- coupling (possibly requiring a multi-
(Class A/B, Class B, Class C, Class F, ment), without further optimization, layer printed circuit board (PCB), or
inverse Class F, etc.) of each amplifi- results in considerable performance a custom component). The approach
er is also important. The two basic sacrifice. Of course, this approach is employed here is realizable with a
considerations for each stage are the predicated on the availability of accu- single layer PCB, and it allows con-
bias conditions (biased on, or pinched rate non-linear transistor models and siderable flexibility in setting the
off, and to what degree) and the fun- CAD tools such as the Microwave coupling ratio.
damental and harmonic impedance Office non-linear (harmonic balance) The Doherty amplifier perfor-
terminations presented to the tran- simulator. mance was simulated for various
sistors. Previous papers on Doherty The potential for correction using input couplers and peaking amplifier
design have advocated particular digital predistortion is improved by bias levels. All simulations are at 2.6
classes of operation for the carrier two additional features: (1) maximiz- GHz with a drain supply voltage of
and peaking amplifiers [6]. The ing the RF bandwidth of the Doherty +28 volts DC. The devices are Cree
approach described in this article was amplifier, and (2) increasing the video CGH27030F GaN HEMT transistors.
to start with a waveform-engineered bandwidth by minimizing the drain The plot of Figure 3 shows the
design, in which the “stand-alone” bias feed inductance. result of simulations for two Doherty
stage (i.e., outside the Doherty envi- amplifiers that are identical except
ronment) was optimized in a 50-ohm Sensitivity to Input Divider for the input couplers. The dark trace
system. Then the biases were adjust- Coupling Factor and Stage Bias is for equal power to both the carrier
ed for carrier and peaking functions The input power divider for the and peaking amplifiers. The light
and the stages were inserted into the design is a variation of a 2-section trace is for an unequal power divider
Doherty configuration. Waveform Wilkinson divider (Fig. 2), with the (approximately 1.8/4.7 dB coupling)
engineering [7] was used to maximize second isolation resistor omitted in which approximately 3 dB more

20 High Frequency Electronics

 High Frequency Design
DOHERTY AMPLIFIER

Figure 4 · Simulated gain and effi-

ciency versus CW output power, for
different gate bias voltages for the
peaking amplifier (1.8/4.7 dB input
coupler). The dark trace is for Vgs =
–5 volts. Voltage steps are 0.5 V.

power is directed to the carrier Figure 5 · Doherty amplifier schematic.

amplifier than to the peaking ampli-
fier. The unequal coupling case sup-
ports significantly higher efficiency examine the effectiveness of digital
in the backoff power range (a 4.3% predistortion for various peaking
efficiency improvement at 10 dB amplifier gate bias levels.
below the peak power is predicted as
well as more uniform gain as a func- Amplifier Schematic and
tion of output power). The Doherty Drain Bias Design
amplifier used for this simulation is The 2.5-2.7 GHz Doherty amplifi-
not strictly an efficiency-optimized er schematic is shown in Figure 5.
amplifier—rather, it was optimized The input and output matching cir-
with equal emphasis on peak power, cuits of the carrier and peaking Figure 6 · Simulated 2-tone inter-
linearity and efficiency. amplifier stages use conventional modulation products versus output
The simulations of Figure 4 are matching circuitry. The open circuit power level (tone separation of 1
for the unequal input divider, with stubs on the input and output are MHz).
the gate bias voltage to the peaking placed at critical junctions where
amplifier varied from –3.0 to –5.0 they have the desired impact on
volts. The more negative the gate source and load impedances. The out- ments) impedance at video frequen-
voltage, the later the peaking ampli- put circuit includes two drain bias cies is also simulated. For this circuit,
fier turns on as power is increased. lines per device, in order to reduce the magnitude of the impedance pre-
These simulations suggest that video impedance. Additionally, the sented to the drain increases mono-
increasing the gate voltage (more length of these bias lines is chosen to tonically from a few milliohms at 100
negative) improves efficiency in back- assist in positioning the load present- kHz to 200 milliohms at 10 MHz, and
off, progressively introduces more ed to the devices at second harmonic approximately 1 ohm at 50 MHz
AM/AM, but—surprisingly—also can frequencies. The impedance at the (dominated by the effective induc-
decrease the AM/PM. The AM/PM is second harmonics is approximately a tance of the bias feeds).
predicted to be very low at the opti- short circuit (as a stand-alone sub- The typical 3rd order two tone
mum gate bias. These simulations circuit in a 50-ohm system), although intermodulation products as a func-
are for CW signals, and the results the precise angle (bias line length) is tion of output power for the Doherty
will differ for dynamic WiMAX sig- adjusted during simulation of the full amplifier design are also examined
nals. An improvement in efficiency Doherty amplifier. as these relate directly to the error
and linearity is observed with The motivation for limiting the vector magnitude (EVM) and relative
increasing gate voltage for the hard- inductance of the drain feed lines is constellation error (RCE) of the
ware under WiMAX drive signals. It to reduce memory effects, which can amplifier when driven by WiMAX
is also worth noting that these plots be exacerbated by bias circuits. The signals. The simulations of Figure 6
(Fig. 4) suggest that it is fruitful to output circuit (including bias ele- demonstrate a typical “back-off hill”

22 High Frequency Electronics

 High Frequency Design
DOHERTY AMPLIFIER

quarter wave line connecting the car-

rier and peaking amplifier outputs is
a sensitive parameter, as are the
open circuit stubs near this line. The
output transformer was found to
exhibit relatively low sensitivity.
The Doherty amplifier is shown
Figure 8. The devices are Cree
CGH27030F GaN HEMT transistors.
The printed circuit board is Rogers
Figure 7 · Simulated 2-tone IM Figure 8 · The 2.5-2.7 GHz Doherty RO4350B material with 0.5 mm
products imbalance versus fre- amplifier with two 30W GaN HEMTs thickness. The amplifier construction
quency separation of tones (Pout = PCB size is 2.5 × 4.0 inches. is conventional, using commonly
+32 dBm). available components and materials.
Only a few of the tuning pads were
used in the prototype alignment, as
the transistor modeling and simula-
tions proved to be accurate.

Time Domain Simulations of

Doherty Operation
The Doherty amplifier was
designed from a frequency domain
perspective, modifying elements to
improve efficiency, 2-tone linearity,
and peak power. This process
Figure 9 · Simulation of drain cur- Figure 10 · Simulation of currents at involved simulations commonly used
rents at the carrier (black) and Doherty output combiner node, sin- by microwave engineers, including
peaking (red) transistors with sin- gle-tone Pout = 48 dBm; black = out- CW gain and efficiency versus output
gle-tone Pout = 36, 42 and 48 dBm. put arm, red = carrier arm, blue = power, and two tone IM distortion
peaking arm. versus output power. However it is
also informative to examine the
in intermodulation distortion at output level (illustrative of the hill resulting amplifier from the time
about +32 dBm average output. Note region). This simulation shows that domain perspective. The plots of
that upper and lower IM products (at the imbalance of 3rd and 5th order Figures 9 and 10 are of currents at
frequencies 2F2–F1 and 2F1–F2) are intermodulation tones (at +32 dBm critical nodes in the circuit (F = 2.6
nearly equal except in the “sweet output) is less than 0.5 dB for tone GHz, Vds = +28V, Idq = 170 mA for the
spot” power range, for this case of a 1 separations to 18 MHz. This is signif- carrier amplifier and Vgs = –4.0 volts
MHz difference between F1 and F2. icant for applications involving 10 for the peaking amplifier).
The imbalance between the upper MHz wide WiMAX signals. The simulations resulting in
and lower IM products is a function The output matching structures of Figure 9 show the drain currents at
of several parameters, including the the carrier and peaking amplifiers the two transistors, for three output
impedances presented to the devices were adjusted while examining both power levels of +36 dBm, +42 dBm
at fundamental, harmonic, sum and the 2-tone IM rejection and the single and +48 dBm. Recall that the drive to
difference frequencies. The bias cir- tone gain and efficiency versus out- the peaking amplifier is offset (rela-
cuit impacts the impedances at the put power. Critical circuit elements tive to that to the carrier amplifier)
difference (video) frequencies. It is included the matching elements at by 90 degrees due to the delay line in
illustrative to simulate the 2-tone IM the fundamental frequency, as well as the Doherty circuitry. The peak tran-
rejection at the hill power (+32 dBm) the bias lines for harmonic matching. sistor currents contribute strongly
versus frequency of separation All of these elements were adjusted only at the highest output power. It is
between the tones. The inequality of after integration into the Doherty interesting to note the relatively non-
IM products versus tone separation configuration (relative to the values sinusoidal shape of the individual
differs depending on power level, and in a stand alone 50-ohm environ- waveforms. The simulated currents
the plot of Figure 7 is for a +32 dBm ment). The width (impedance) of the are at the die drain pads, and the

24 High Frequency Electronics

Figure 12 · Measured Doherty amplifier EVM and effi- Figure 13 · Measured Doherty amplifier ACPR and effi-
ciency vs. WiMAX Pout (without predistortion). ciency vs. WiMAX Pout (without predistortion).

intrinsic capacitances of these ly sinusoidal waveform. It is interest- GHz, not 2.6 GHz) would offer the
devices are clearly important. These ing to observe that the role of the potential for incrementally higher
waveforms suggest some sort of Class peaking amplifier appears to be more efficiency.
J operation as defined by Cripps [7]. a matter of complex waveform syn-
These current simulations show that thesis, as opposed to the more ele- Description of WiMAX Protocol
the higher order (3rd, 4th, etc.) har- mentary view involving Class C like and Digital Predistorter Design
monic terminations on the amplifier current pulses. The digital predistortion (DPD)
outputs are non-ideal for efficiency test-bed used for correction of this
optimization since the waveforms are Measured Amplifier Performance Doherty amplifier was provided by
not “squared off,” but are rather soft The design was optimized for lin- Optichron. The test-bed used is the
in their transitions. This is due to the earity and efficiency over the 2.5-2.7 base-band development board with
emphasis on (uncorrected) linearity GHz range to support a number of the OP4400 chipset. A block diagram
during circuit optimization. possible WiMAX frequencies. The of the system is shown in Figure 14.
The currents shown in Figure 10 small signal bandwidth exceeds this The WiMAX signal, which is a fully
are those flowing into the Doherty limited range as shown in Figure 11. compliant 802.16e-2005 signal, is
combining node for the main and Gain is nominally 13 dB and input used as a stimulus to the test-bed.
peaking amplifiers, and exiting the return loss is typically 14 dB. This signal is generated using Rohde
node for the output arm at +48 dBm The efficiency and linearity of the and Schwarz AMIQ equipment and
output power. The output arm cur- final amplifier were evaluated using a applied to the test-bed in digital I and
rent equals the sum of the main and 10 MHz wide WiMAX signal compli- Q format. When setting up the test-
peaking arm currents. The currents ant to 802.16e-2005. The key linearity bed the signal is first passed through
add constructively to produce a near- parameters for this modulation the OP4400 chipset with no correc-
scheme are error vector magnitude tion applied. This allows the perfor-
(EVM), as shown in Figure 12, and mance of the uncorrected amplifier to
spectral emissions mask (SEM). The be observed. The digital signal is fed
spectral mask was measured using a to a Texas Instruments TSW3003
10 MHz wide integration bandwidth upconverter. The up-converted signal
in the adjacent channel, as shown in level at the output of the radio board
Figure 13, at 2.5, 2.6 and 2.7 GHz. is approximately –10 dBm and there-
The performance is optimum at 2.6 fore needs to be amplified to be able
GHz with trade-offs in linearity at the to drive the Doherty amplifier to the
band edges. Efficiency versus output required power levels. It should be
Figure 11 · Measured small-signal power is flat over 2.5 to 2.6 GHz drop- noted that this amplifier is located
gain and input and output return ping by approximately 5% at 2.7 GHz. within the predistortion loop (labeled
loss versus frequency (output return It is clear that single frequency in Figure 14 as the “Driving Amp”),
loss is dashed line). designs (optimized at either 2.5 or 2.7 which means that it’s performance,

December 2008 25
 High Frequency Design
DOHERTY AMPLIFIER

fier down to the system noise floor

associated with the 50 dB gain driver
amplifier. This was somewhat com-
pensated with the use of attenuation
at the output of that amplifier. At the
maximum output power both algo-
rithms were needed due to the large
amounts of clipping occurring at the
peaks of the WiMAX signal. At 8
watts average output power the
crests were being clipped from their
theoretical level of 100 watts by the
peak capability of the Doherty ampli-
fier which was estimated to be about
60 watts. This equates to more than 2
dB compression assuming that the
peak-to-average power ratio (PAPR)
was not increased by the digital pre-
Figure 14 · Digital predistortion test-bed using Optichron OP4400 chipset. distortion algorithm!
The required SEM at 1.5 MHz off-
set from the edge of the signal is –45
albeit extremely linear, is also being point a digitally predistorted signal is dB. This specification was met at 8
corrected by the test-bed. The result- then reapplied to the amplifier under watts of average output power and at
ing modulated signal is then applied test in real time. The option exists 47% DC-RF efficiency for a WiMAX
to the amplifier under test which cre- within the test-bed’s software con- 10 MHz channel bandwidth signal
ates some level of unwanted distor- trols for the user to decide whether to with 11 dB peak-to-average ratio, as
tion. The output signal is sampled use a simple memoryless algorithm shown in Figure 15 . The correction in
and downconverted using a commer- or whether to choose a more complex SEM with the Optichron OP4400
cially available Mini Circuits mixer. algorithm which would help correct DPD test-bed was 17 dB. The plot of
The IF signal is then processed by an amplifiers with strong memory Figure 16 shows that such excellent
Analog Devices analog to digital con- effects. correction is maintained across fre-
verter (AD9233). The resulting digi- It was found that, at low average quency (2.5 and 2.655 GHz) as well
tal representation of the output sig- power levels, the memoryless algo- as with average output power. It can
nal is then processed by the rithm was sufficient to correct the be seen that the distortion products
Optichron OP4400 chipset. At this non-linearities of the Doherty ampli- of a Doherty amplifier are relatively

Figure 15 · Doherty amplifier spectral mask, with and Figure 16 · Doherty amplifier spectral mask vs. aver-
without DPD correction (F = 2.5 GHz, Pave = 8 watts, age output power, with and without DPD correction (F =
Vsupply = +28 V, I = 600 mA, 48% DC-RF efficiency). 2.5, 2.655 GHz, Vsupply = +28 V).

26 High Frequency Electronics

 High Frequency Design
DOHERTY AMPLIFIER

high when the output power is posium Digest, 2004, pp. 529-532. fessional activities, he has written
backed off quite heavily from the 4. M.J. Pelk, W.D.E. Neo, J. numerous magazine articles and has
maximum average output power. The Gajadharsing, R. Pengelly, L.C.N. de presented papers and led workshops
DPD system, however, is able to cor- Vreede, “A High Efficiency 100W at international conferences. He
rect down to the system noise floor GaN Doherty Amplifier for Base- served as secretary on the steering
with better than 20 dB correction Station Applications,” IEEE committee for IMS2006 in San
over a wide dynamic range. Transactions on Microwave Theory Francisco and is currently acting on
These results demonstrate the and Techniques, July 2008, pp. 1582- the RWS Technical Paper Committee.
feasibility of a compact, high efficien- 1591. He holds five United States patents
cy design for WiMAX transmitters at 5. U. Andre, J. Crescenzi, R. in amplifier design. Simon can be
2.5 or 2.7 GHz with up to 8 watts Pengelly, A. Prejs, S. Wood, “High reached at Simon_Wood@cree.com
average output power. Efficiency, High Linearity GaN Ray Pengelly gained his BSc. and
HEMT Amplifiers for WiMAX MSc. degrees from Southampton
Summary and Conclusions Applications,” High Frequency University, England in 1969 and
A number of design considera- Electronics, June 2007, Vol. 6, No. 6., 1973, respectively. Since August
tions for Doherty amplifiers using pp. 16-29. 1999, Ray has been employed by Cree
GaN HEMT transistors in the 2.5-2.7 6. J. Kim, B. Kim, Y. Yun Woo, Inc. Initially, he was the General
GHz bands have been presented. The “Advanced Design of Linear Doherty Manager for Cree Microwave respon-
critical roles of the input coupler and Amplifier for High Efficiency using sible for bringing Cree’s wide
peaking amplifier bias point were Saturation Amplifier,” 2007 IEEE bandgap transistor technology to the
described, as well as the rationale for MTT-S International Symposium commercial marketplace. On the
input divider design and the choice of Digest, pp. 1573-1576. acquisition of UltraRF, Sunnyvale in
amplifier bias. The importance of bias 7. S.C. Cripps, RF Power 2001, he became Chief Technical
circuit design and the evaluation of Amplifiers for Wireless Communi- Officer enabling his skills in
such circuits were also covered. The cations, 2nd edition, Artech House, microwave semiconductor technology,
design optimization process was 2006, pp. 68-73, 122-124. based on 30 years of experience, to be
described, including the key parame- 8. R.S. Pengelly, “N-way RF power utilized. From September 2005 he
ters and tradeoffs involved. Finally, amplifier with increased backoff became responsible for new business
the effectiveness of digital predistor- power and power added efficiency,” development of wide bandgap tech-
tion for this Doherty amplifier with U.S. Patent 6 700 444, Mar. 2, 2004. nologies for RF and microwave appli-
GaN HEMT transistors was demon- 9. R.S. Pengelly and S.M. Wood, cations for Cree in Durham, NC and
strated. The result was a fully speci- “N-way RF power amplifier circuit most recently has been involved with
fication compliant WiMAX amplifier with increased back-off capability the release of GaN HEMT transistors
with 8 watts average output level at and power-added efficiency using and ICs for general purpose and
an extraordinary 47% efficiency. unequal input power division,” U.S. telecommunications applications.
Patent 6,737,922, May 18, 2004. Ray has written over 100 technical
References: 10. R.S. Pengelly and S.M. Wood, papers, 4 technical books, holds 12
1. S.C. Cripps, RF Power “N-way RF power amplifier circuit patents and is a Fellow of the IET
Amplifiers for Wireless Communi- with increased back-off capability and Senior Member of the IEEE.
cations, 2nd edition, Artech House, and power-added efficiency using Jim Crescenzi received the BS
2006, pp. 290-303. selected phase lengths and output degree at UC Berkeley in 1961 and
2. J. Kim, J. Cha, I. Kim, and B. impedances,” U.S. Patent 6,791,417, the MS and PhD degrees at the
Kim, “Optimum operation of asym- Sep. 14, 2004. University of Colorado in 1962 and
metrical cell-based linear Doherty 1969. He founded Central Coast
power amplifiers—uneven power Author Information Microwave Design in 2005, where he
drive and power matching,” IEEE Simon Wood has a Bachelor of provides consulting services primari-
Trans. Microwave Theory & Engineering degree from the ly for microwave power amplifiers
Techniques, vol. 53, no. 5, May 2005, University of Bradford, UK. Since and subsystems. Jim is a Life Fellow
pp. 1802-1809. joining Cree Inc. in 2000, he has of the IEEE for contributions to the
3. J. Gajadharsing, O. Bosma, and designed amplifiers using SiC MES- development of microwave ampli-
P. Van Western, “Analysis and design FET, Si LDMOS, and more recently fiers, integrated circuit technology,
of a 200W LDMOS based Doherty GaN HEMT devices. He has been and miniature receivers for defense
amplifier for 3-G base station,” IEEE Manager of Product Development at applications. He can be reached at
MTT-S International Microwave Sym- Cree since January 2006. In his pro- jcrescenzi@gmail.com.

28 High Frequency Electronics