The Maximally Efficient Amplifier - Gareth Lloyd Rohde & Schwarz, Munich, Germany - Rohde & Schwarz
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
The Maximally Efficient
Amplifier
Gareth Lloyd
Rohde & Schwarz, Munich, Germany
T
he energy efficiency of an RF front- for improving amplifier efficiency are clas-
end (RFFE) is a vital characteristic, sified by their mechanisms—also noting
whether a radio is battery or mains the mechanisms not used—and identifying
powered. For battery powered, re- areas for improvement. Finally, the article
ducing the maximum current drawn from the shows that harmonic load-pull measure-
battery increases the time between charges. ments on a device highlight its potential; us-
For mains powered, important properties ing such measurement data with a look-up
such as size, weight and power are dictated table, for example, amplifier performance in
by the RFFE efficiency. Consequently, many a variety of schemes can be predicted.
amplifier architectures and inventions have
been developed to minimize wasted energy THE TUNED AMPLIFIER
in the transmitter. Although improving effi- A tuned amplifier circuit can be used to
ciency, some of these rely on theoretically describe the continuum of amplifier class
impossible modes of opera- characteristics from A to C, via AB and B,
Vsupply tion, and some fail to fully use based on sinusoidal voltage waveforms and
the device’s capabilities. quasi-linear operation. A more detailed ex-
Lowpass Bandpass This article provides some planation is provided in Chapter 3 of Cripps’
Filter Filter analysis of and insight into book.1 A simplified model of the power
amplifier efficiency. First, the amplifier built around a controlled current
tuned amplifier concept and source is shown in Figure 1. The model can
(Vd, id) Highpass Zload efficiency enhancing mecha- be simplified into three frequency domains:
Filter nisms are explained. Then, the • DC current flows through the lowpass fil-
effects of each mechanism on ter and the controlled current source (i.e.,
amplifier efficiency are illus- the device). Its progress elsewhere in the
Fig. 1 Simplified schematic of a trated, revealing some surpris- circuit is blocked by the bandpass and
tuned amplifier, class A to C. es. The better known methods highpass filters.
Reprinted with permission of MICROWAVE JOURNAL® from the April 2020 issue.
©2020 Horizon House Publications, Inc.
TechnicalFeature
• At the fundamental frequency, the components and is sinusoidal. Pow- the mechanisms for reducing the
signal current through the device er is dissipated by the device when wasted power in a tuned amplifier.
passes solely to the intended load a current (id) flows through it with a These mechanisms relate only to the
impedance (Zload), creating a volt- voltage (vd) across it and where the device itself, not to external modu-
age across the device and load. current and voltage overlap during lating circuitry such as harmonic
• Harmonic currents flowing the waveform cycle. For a class A terminations or modulators. Three
through the device are short amplifier, the simplest case, Figure 2 base mechanisms can enhance the
circuited through the highpass shows the power dissipation versus efficiency of a single-ended ampli-
filter, as any harmonic currents time at three power levels. As the fier: waveform engineering, supply
flowing in the device “see” zero output power reduces, the power modulation and load modulation.
impedance and do not create dissipated waveform tends to a con- Waveform Engineering—The
any voltage. stant value. At higher output power, shape of the voltage and/or current
The voltage across the device the dissipated power reduces. The waveform is modified, which is what
comprises only DC and fundamental power consumption is constant in all happens when passing through the
cases, and the pow- class A to C continuum. Harmonic
Current Through
er dissipated is the content is introduced into the cur-
2 2 2 total area under the rent, modifying its waveform, in a
1 1 1
power dissipation predictable but restricted way.1 Al-
curve. In the case of ternatively, the ratio of the current’s
1 this class A ampli- harmonic content may be modified
0 0 0
0 0.5 1 0 0.5 1 0 0.5
fier, the amount of by injecting harmonics from either
Voltage Across
power dissipated the input side or output side. For the
2 2 2
(wasted) decreases current’s harmonic content to affect
1 1 1 as output power the voltage waveform, a non-zero im-
0 0
level increases, pedance must be present at that har-
0
0 0.5 1 0 0.5 1 0 0.5 1 from (a) to (c). monic frequency. In the limiting case,
both current and voltage waveforms
Power Dissipation
2 2 2
Instantaneous
EFFICIENCY are square waves and antiphase. As
1 1 1 ENHANCEMENT one of them is zero at any instant in
How will effi- time, the power dissipation is zero.
0 0 0 ciency enhance- This zero dissipation applies at least
0 0.5 1 0 0.5 1 0 0.5 1
ment mechanisms to the device, but it could just be
(a) Time (b) Time (c) Time
improve the ener- shifted elsewhere in the system.
gy efficiency? Con- Supply Modulation—The aver-
Fig. 2 Class A amplifier waveforms at low (a), medium (b)
sider classifying age or envelope supply voltage
and high (c) output power.
across the device, Vsupply, is modi-
fied. With a perfect device, Vsupply
is the root-mean-square value of the
Current Through
2 2 2 2
voltage waveform, set so the mini-
mum value of vd reaches zero.
1 1 1 1
Load Modulation—The Zload
presented to the device at the fun-
0 0 0 0
0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1 damental frequency is modified,
ideally so the voltage (vd) swings
from 0 to 2 times the supply.
Voltage Across
2 2 2 2 Figure 3 illustrates these mecha-
nisms using a class B waveform as
1 1 1 1 the reference. The class B current
waveform is half sinusoidal, com-
0 0 0 0 pared to fully sinusoidal for class
0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1
A operation. The fundamental cur-
1 1 1 1 rent is equal in both classes A and
B when the peak-to-peak current
Power Dissipation
Instantaneous
variation is equal. The reason for us-
0.5 0.5 0.5 0.5 ing a class B waveform as the base-
line is because the efficiency can
be enhanced by all three methods.
0
0 0.5 1
0
0 0.5 1
0
0 0.5 1
0
0 0.5 1
Class A, on the other hand, cannot
(a) Time (b) Time (c) Time (d) Time
be improved with load modulation
alone: class A power consumption
Fig. 3 Efficiency enhancement mechanisms compared to a class B waveform at 6 dB remains constant regardless of the
back-off (a), waveform engineering (b), supply modulation (c) and load modulation (d). load impedance.
TechnicalFeature
ply modulation has been applied—
achieving a quite unexpected result:
Current Through
2 2 2 2 2 the efficiency at the output power
back-off of the class A amplifier has
1 1 1 1 1 been maintained at the theoretical
maximum of 50 percent.
0
0 0.5 1
0
0 0.5 1
0
0 0.5 1
0
0 0.5 1
0
0 0.5 1
The respective efficiencies of the
five scenarios of Figure 4 are sum-
marized in Table 1. Note that the
Voltage Across
waveforms all have the same output
2 2 2 2 2
power.
1 1 1 1 1
POPULAR APPROACHES
0 0 0 0 0
Having classified various en-
0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1 0 0.5 1 hancement mechanisms and dis-
3 3 3 3 3
cussed their effects on theoretical
amplifier blocks, including advan-
Power Dissipation
tageous effects from hybrid ap-
Instantaneous
2 2 2 2 2
proaches, the discussion moves
1 1 1 1 1 from theory to practice, classifying
the popular enhancement methods
0
0 0.5 1
0
0 0.5 1
0
0 0.5 1
0
0 0.5 1
0
0 0.5 1
according to the mechanisms they
(a) Time (b) Time (c) Time (d) Time (e) Time
use (see Figure 5). Using a Venn
diagram for classification helps
identify where additional schemes
Fig. 4 Class A amplifier efficiency at 6 dB back-off (a) enhanced by waveform
are complementary and may further
engineering (b), supply modulation (c), load modulation (d) and an unexpected
improvement using a counter-intuitive hybrid (e). improve efficiency.
For example, the Doherty ampli-
waveform engi-
TABLE 1 neered to add the
fier, which applies load modulation
to its constituent amplifiers, can be
APPROACHES TO INCREASE CLASS A required harmon-
AMPLIFIER EFFICIENCY improved by adding supply modu-
ics? For example, it lation, especially to the main chan-
% Efficiency at 6 dB Back-Off might be operating nel, and/or waveform engineering,
Configuration close to its upper
(25% of Rated Power) by modifying the design to incor-
frequency limit and porate class F-1 operation, for ex-
Class A Baseline 12.5
cannot support the ample.
Waveform Engineered harmonic currents.
39.3
(Equivalent to Class B) Supply modulation Harmonic Load-Pull
Supply Modulation 25.0 in Fig. 4(c) could A bottleneck is getting real world,
be used, although practical devices to use the theoret-
Load Modulation 12.5 combining it with ical enhancements. For example, a
Anti-Load & Supply Modulation 50.0 load modulation typical GaN device may be sensitive
would be coun- to efficiency enhancement by load
terproductive. Load modulation in- modulation over a 5–10:1 imped-
Class A Case Study creases the peak-to-peak voltage, ance range. However, when used as
One of the goals of this article decreasing the range where sup- the main device in a Doherty archi-
is to illustrate efficiency enhance- ply modulation could be deployed. tecture, it is typically exploited only
ment mechanisms so they can be Turning the problem around, if load over a 2–3:1 range. The Doherty
used optimally. The class A case is modulation degrades the effective- scheme will fail to maximize the po-
not a lost cause. Figure 4 shows the ness of supply modulation, what if tential performance of the device.
voltage, current and dissipation for “unload” modulation were applied? Harmonic load-pull measure-
a class A amplifier, illustrating that Instead of maximizing the load im- ments over a range of bias condi-
waveform engineering and supply pedance to maximize peak-to-peak tions make it possible to establish
modulation enhance efficiency, but voltage, minimize the load imped- the maximum performance enve-
load modulation does not. Wave- ance to minimize the peak-to-peak lope for the device technology.
form engineering can convert the voltage and then use supply modu- Load-pull data can be obtained us-
class A sinusoidal current into the lation. This is the case shown in Fig- ing various setups, such as Maury
class B case of the half sinusoid ure 4(e), the “anti-load modulation + Microwave’s harmonic load-pull
in Fig. 4(b). Referring to Figure 3, supply modulation” case. The peak- test bench with an R&S ZNA vector
class B efficiency could now be en- to-peak values of current and voltage network analyzer (see Figure 6). By
hanced with load modulation. have been completely reversed from comparing harmonic load-pull mea-
What if the device could not be the load modulation case, and sup- surement data with the theoretical
TechnicalFeature
power approaches 15 dB.
Load Supply For a Doherty amplifier, the (sim-
Doherty-Main
Modulation Modulation plified) relationship between output
Doherty-
Auxiliary Multilevel power and impedance is defined by:
Outphasing
Outphasing Envelope iaux = 2imain-1, where imain, the nor-
Tracking (ET)
Load malized output current from the
Modulation
Balanced
main transistor, varies from 0 to 1.
iaux = 0 where iaux < 0. iaux is the nor-
Doherty Class malized output current of the auxil-
+ Class F-1 F + ET iary device.
Class F
The normalized impedances pre-
sented to the main and auxiliary
Switched-Mode transistors are
Power
Amplifiers Zmain = 2 – iaux/imain and Zaux = 1/
(iaux/imain), respectively.
Waveform
Engineering The output power contributions
from the constituent amplifiers are
Fig. 5 Efficiency enhancement mechanisms, their hybrids and possible areas for
given by
further improvement.
Pmain = imain2 × Zmain and Paux =iaux2
Thus there are several ways to use × Zaux
the data from a rigorous and repeat-
able setup for measuring load-pull. In this Doherty example, a the-
One is to create characterization and oretical output current relation-
architectural databases for cross- ship is used, although the equa-
referencing device performance with tion relating iaux and imain can be
various circuit architectures and changed, using a square law auxil-
enhancement methods: Doherty, iary relationship for example, where
load modulated, balanced, outphas- iaux=√imain. The impedance values
ing, etc. The design flow of Figure Zmain and Zaux may be scaled to any
Fig. 6 Maury load-pull test system 7 shows possible steps for setting value in the dataset. In this case,
using the R&S®ZNA VNA. 10 Ω is used as a multiplier prior to
up and using a look-up table for
assessing device performance and the look-up operation; fixed values
performance of a selected high ef- enhancement techniques. To illus- for the harmonic impedances have
ficiency technique, the performance trate the concept, a commercially also been chosen prior to look-up.
gap can be quantified, answering available Wolfspeed GaN transis- With the definition for output pow-
the question of the difference be- tor (CG2H40010) was characterized er and impedance for the main and
tween what has been built and the at a fundamental frequency of 2 auxiliary devices, the look-up-table
performance limit. Alternatively, GHz and a bias current of 100 mA operation is performed on the mea-
if the device is assumed to be the using a Maury harmonic load-pull surement data to extract the drain
bottleneck, the harmonic load- test bench. The measurement data efficiency, with interpolation of the
pull measurement data enables a was analyzed to understand device measurements used to determine
scheme to be designed to maximize performance in a Doherty amplifier, intermediate values.
its potential, using the optimal en- then compared with the maximum With the output power and drain
hancement mechanisms in the cor- performance possible from the de- efficiency for the main and auxiliary
rect proportions. vice. Figure 8 shows the output known individually, the compos-
power and drain ef- ite power consumption and out-
ficiency as the input put power can be calculated. The
Perform Harmonic Load-Pull
(Efficiency Pout) = f(Vsupply, Zf0, Zfn)
Create Database of Architectures
power, fundamental simulated output power and drain
(Zf0, Zfn, Vsupply) = f(Pout)
and harmonic ter- efficiency of the Doherty power am-
minations and sup- plifier is plotted in Figure 9, using
Add to Device Database Add to Architecture Database
ply voltages were the measured data for the look-up
swept. This scatter operation. Because two devices are
Use Lookup Table to Extract Device Performances in Different plot provides the used in a Doherty, the output power
Architectures (Doherty, Outphasing, Class F, ET)
performance limit capability is 3 dB higher, so a second
of a single-ended scatter plot of the measured data,
Compare Opportunity vs. Cost. Select Device and Application increased by 3 dB, is included. The
device; to achieve
drain efficiency second scatter plot represents the
Implement greater than 50 per- performance limit. The load modu-
cent, the dynamic lation mechanism offered by the
Fig. 7 Process flow to achieve the best device-architecture range of the output Doherty architecture—the limited
pairing.TechnicalFeature
90 90
80 80
70 70
Drain Efficiency (%)
60
Drain Efficiency (%)
60
50 50
40 40
30 30
20 20
• Fixed (Znf0, Vds)
10 10 • Fixed (Znf0, Vds) + 3 dB
Doherty
0 0
25 30 35 40 45 25 30 35 40 45
Output Power (dBm) Output Power (dBm)
Fig. 8 Single-ended GaN transistor drain efficiency and
Fig. 9 Calculated drain efficiency and output power vs.
measured drain efficiency and output power with fixed supply
output power vs. swept input power, supply voltage (10, voltage and harmonic impedances.
20 and 28 V) and fundamental, second and third harmonic
impedances.
This model for the Doherty could be more sophis-
90
ticated, including other effects without detracting from
80 the basic flow. Alternatively, it could be that a different
enhancement scheme offers a greater benefit for the
70 device, whether tailored from the ground up or off-the-
shelf. Other concepts can be analyzed using different
Drain Efficiency (%)
60
equations and look-up parameters. For example, using
50 the outphasing architecture, the impedances presented
40
to the voltage source devices are derived from the co-
tangent of the inverse sine of the output amplitude.1
30
CONCLUSION
20
A classification of efficiency enhancement mecha-
10 • Measured Possible nisms has been proposed, and their effects on class A
Calculated Doherty and class B amplifiers described, allowing for comple-
0 mentary mechanisms to be identified.
25 30 35 40 45
Output Power (dBm) It is suggested that harmonic load-pull measure-
ments, over a range of supply voltages, which are anal-
Fig. 10 Calculated drain efficiency and output power vs. ogous to the mechanisms, can fully extract the perfor-
measured drain efficiency and output power showing full mance potential of the device.
supply voltage and harmonic impedance variations. From those measurements, performance of the de-
2:1 modulation range presented to the dominant main vice in a range of architectures (e.g. Doherty) can be
device, combined with the arbitrary impedance trajec- predicted.
tory selected for this illustration—does not fully exploit State-of-the-art devices, such as the Wolfspeed de-
the device’s capability for load modulation. The device vice illustrated in this article, are capable of much better
is, in effect, being driven in first gear. While the measure- performance than state-of-the-art architectures.
ment space indicates a capability of 8 dB dynamic range Therefore, that designing a “good enough” supply-
achieving at least 50 percent drain efficiency, the Doherty modulated harmonic load-pull, appropriate for the ap-
only manages to exploit about 5 dB of that range, also plication at hand, should be a goal for those responsible
missing a couple of dB of saturated output power. The for developing power efficient RFFEs.n
same extracted Doherty performance plotted on the en-
ACKNOWLEDGMENT
tire measurement space, including the full harmonic and
supply variations, is shown in Figure 10. For efficiency of The author would like to thank Maury Microwave for
50 percent or greater, the output power dynamic range providing the measurement data.
is now nearly 18 dB with the addition of the auxiliary tran-
Reference
sistor’s 3 dB contribution. Clearly the performance of the 1. Cripps, S. C., “RF Power Amplifiers for Wireless Communica-
Doherty in this example would benefit from the addition tions,” 2006, Artech House, Norwood, Mass., Chapter 3.
of supply modulation and/or waveform engineering.You can also read
