Supply modulation study for Ericsson MINI LINK

Supply modulation study for Ericsson MINI LINK
Supply modulation study for Ericsson MINI LINK
Master of Science Thesis in Wireless and Photonics Engineering
BJÖRN HERRMANN
Department of Microtechnology and Nanosciency
Division of Microwave Electronics Laboratory
CHALMERS UNIVERSITY OF TECHNOLOGY
Gothenburg, Sweden, 2011
Abstract
This report presents a feasibility study of envelope tracking for Ericssons radio
link system. The analysis is done based on a model of an existing 7 GHz GaAs
Power amplifier. From the on the model based analysis the theoretical PAE was
found to be 50% compared to the 15% achieved through static biasing.
The high capacity radio links which were the basis of this study requires
bandwidths up to 100 MHz. Implementing envelope tracking on such a systems
introduces problems both for linearization and supply modulation efficiency. By
introducing an iterative peak tracking low pass this report shows that the PAE still
can be improved by at least 5% with a low efficiency supply modulator. Widening
the study and investigating the effect of modulating the supply of the whole
amplifier chain showed a potential improvement of 10%
The trade off between, reduced PAE improvement in the PA and efficiency
improvement in the supply modulator when reducing the bandwidth reduction,
was evaluated by simulating a simple envelope amplifier. Using lumped elements
the maximum bandwidth achieved was 40 MHz.
Table of Contents
Abstract .................................................................................................. 1
Table of Contents................................................................................... 2
1
Abbreviations............................................................................. 3
2
Introduction ............................................................................... 4
2.1
2.2
Dynamic supply modulation ................................................................ 5
Ericsson MINI LINK system ................................................................. 7
3
Characterization of Power Amplifier .......................................... 8
3.1
3.2
3.3
3.3.1
3.3.2
3.3.3
Calculation of supply modulation signals........................................... 11
Input signal calculation ...................................................................... 12
Analysis of PAE improvement potential............................................. 13
Dynamic drain bias modulation PAE improvement potential ............. 13
Dynamic gate bias modulation PAE improvement potential .............. 15
Switched drain bias modulation PAE improvement potential ............ 15
4
Peak tracking bandwidth reduction.......................................... 17
4.1.1
PAE vs EA bandwidth trade off ......................................................... 18
5
Linearity evaluation ................................................................. 19
5.1
5.2
DPD performance .............................................................................. 19
Input spectrum ................................................................................... 21
6
Verification of PAE potential calculations ................................ 22
6.1
Verification of PAE improvement calculations ................................... 22
7
8
Complete PA chain PAE ......................................................... 24
Envelope amplifier ................................................................... 26
8.1
8.2
8.2.1
8.3
8.4
Implementation .................................................................................. 26
Choosing components ....................................................................... 29
Components ...................................................................................... 29
Simulations ........................................................................................ 30
EA simulation results ......................................................................... 30
9
10
11
Conclusions ............................................................................. 33
Future work ............................................................................. 34
References .............................................................................. 35
1
Abbreviations
PA
(Power Amplifier)
PAE
(Power Added efficiency)
DLM
(Dynamic Load Modulation)
DSM
(Dynamic Supply Modulation).
ADS
(Advanced Design System)
DPD
(Digital predistortion)
QAM
(Quadrature amplitude modulation)
PAR
(Peak to average ratio)
DAC
(Digital to analog converter)
RF
(Radio Frequency)
DC
(Direct current)
MSE
(Mean Squared Error)
FET
(Field Effect Transistor)
EER
(Envelope elimination and restoration)
ET
(Envelope tracking)
EA
(Envelope amplifier)
2DLUT
(2 dimensional look up table)
CW
(Constant Wave)
LPF
(Low Pass Filter)
DAC
(Digital to analog converter)
2
Introduction
In all electronic devices the need for managing power consumption is ever
present. Two important factors contributing to this demand are the cost and heat
dissipation associated with high energy consumption. An important parameter in
determining the maximum distance between two radio links in the transmitter
output power, which in turn is limited by internal heat dissipation in the PA (Power
Amplifier). By increasing the efficiency of the PA it is possible to increase radio
link separation and or reduce the power consumption.
The history of PA development is rich in various methods for increasing the PAE
(Power Added Efficiency), Most of these methods are focused on manipulating
the gate bias and recovering the signal through filtering in the load network. In
recent years state of the art performance has been achieved using Doherty, DLM
(Dynamic Load Modulation) or DSM (Dynamic Supply Modulation).
The scope of this project is focused on investigating the feasibility of introducing
the efficiency wise very promising DSM.
The aim of this project is to through simulations and numerical calculations to
evaluate the feasibility and potential of introducing DSM in the radio link units
manufactured by Ericsson.
Conceptually DSM is very simple where it seeks to dynamically change the
transistor supply voltage to the PAE optimal level. What this means in practice is
that for low output power we would like to lower the DC supply voltage and
increase it for higher output power. By not having an unnecessary high bias
voltage the power dissipation is decreased.
As a scheme for optimizing the PAE of power amplifiers, supply modulation has
recently become recognized as a strong candidate for the next generation
wireless standards [1].
2.1
Dynamic supply modulation
When designing a statically biased PA the bias voltages are set to manage the
highest output powers. In order to support the highest output powers the bias
current is unnecessarily high for signals below the peaks. This wasted energy is
the theoretical background behind the concepts of EER (Envelope elimination
and restoration) and ET (Envelope tracking), where the supply bias is controlled
by the envelope of the RF signal via an EA (Envelope amplifier) as can be seen
in Figure 1.
EER means separating the phase and amplitude information in the RF signal.
The phase modulated, constant amplitude signal is then sent into the PA while
the amplitude information is encoded through supply modulation. In ET the input
RF signal is also amplitude modulated. It is this input amplitude which is being
measured and fed to the EA supply modulator.
EA
RF input
PA
RF output
Figure 1. Envelope tracking architecture.
In this project it is assumed that the baseband signal is computed digitally using
an ASIC (Application Specific Integrated Circuit). Digital signal processing allows
for a nonlinear mapping of signal amplitude to supply signal. This in turn enables
continuously driving the amplifier at its maximum PAE. In addition the digital
processing of supply signals means the supply signal level can be fed to the
predistorter. See Figure 2 for a description of the envelope tracking system
considered in this work.
Figure 2. Basic concept of digitally controlled Envelope tracking.
From this general concept three specific cases have been chosen for evaluation.
The feasibility of implementing them on the specific requirements of the Ericsson
radio link system will be presented. The concepts are: Dynamic Gate Bias
Modulation, Dynamic Drain Bias Modulation and Switched Drain Bias Modulation.
They have different advantages and possibilities in terms of feasibility and
potential PAE improvement.
From a PAE perspective dynamic drain bias modulation is the best solution. By
directly controlling the drain voltage the dissipated power can be significantly
reduced. Switching between discrete levels is merely a simplification where the
deviation from the dynamic case is limited by the number of levels and switch
speed. In Figure 3 a time domain description of a switched and continuous drain
modulation is seen. The origin of this envelope signal is a QAM-1024 modulated
carrier. High orders of modulations have implications for the probability
distribution, where the highest peaks become increasingly rare. For such signals
switched solutions could have an increased potential. Another simplification from
the full optimum signal of Figure 3 is a peak tracking bandwidth reduced version.
In this case of course the bandwidth has to be traded for PAE improvement.
Figure 3. Example of drain voltage signal comparing optimum dynamic drain
bias modulation and a switched taking rise and fall time into account.
For dynamic gate bias modulation the bias current for low RF input power is
lowered. This might have implications on transistor gain, which have to be
compensated for by predistortion. The drain current control is not as strong as
when modulation the drain voltage directly, this lowers the maximum PAE
improvement potential. The up side with gate modulation though is that the DC
bias current entering the gate is virtually zero and thus significantly simplifies the
modulator implementation compared to envelope tracking through the drain
supply voltage.
2.2
Ericsson MINI LINK system
This study will be conducted by evaluating DSM applied to a system that exhibits
the very unique requirements of a Radio Link system. Specifically a well
behaving ADS (Advanced Design System) model of a 7GHz PA will be the
subject of the study. This PA is designed to operate in class AB operation with
Idquiescent set at 15.5% of Idss. At 7 GHz this amplifier has a gain of 10dB and a
saturated output power of 39 dBm with a gain saturation of 5 dB.
The DPD (Digital predistorter) currently employed for the 7 GHz PA uses
polynomials which are the inverse of the system. This DPD can not handle gain
expansion and high orders of intermodulation. The systems on which DSM could
be employed will most likely need a more sophisticated DPD because of the non
linearity presented when the bias varies. With the present polynomial DPD the
PA can support a 28dBm average output power and 9dB PAR (Peak to average).
For this study it was decided that the averaged output power was going to be set
to 30 dBm, since efficiency is increased for higher output powers. The PAE
improvement to be evaluated is to be compared with the static performance at
which the PAE is 15%.
Driving the amplifier beyond the point of compression of course means a loss of
gain. This is compensated for by the DPD, the DPD can compensate for a
maximum of 6 dB loss of gain. Allowing for higher DPD swing would require a
DAC with larger dynamic range which in turn would have negative efficiency
implications. This thereby becomes the maximum allowed system gain
compression.
The back haul solutions for transporting information between base stations and
network hubs require very high data rates. Achieving high data rates is often
times done using high orders of modulation as well as a high symbol rates. In
present Ericsson radios QAM (Quadrature amplitude modulation) signals with a
modulation order up to 256 are used. With ever growing demands on link
capacity these requirements will continue to increase. For that reason this study
has been done with a 1024-QAM signal. The symbol rate was chosen to be 85%
of the 112MHz channel spacing in question which is 95.2MHz.
An important signal characteristic is the PAR (Peak to average ratio). It is usually
specified in dB and sets the maximum peak output power in relation to the
average output power. In this study a 9 dB PAR limit was chosen, this because it
was assumed to be close to what can be expected in future MINI LINK systems.
A study of current research literature in envelope amplifiers concluded that the
highest bandwidth achieved is 20MHz [2]. In Figure 4 the spectrum of the QAM1024 signal can be seen. The signal envelope does not have a well defined
bandwidth but could be estimated to around 100MHz. The conclusion being that
either the EA bandwidth has to be increased or this signal bandwidth reduced.
This project aims to do both of those, while investigating interesting trade offs in
terms of PAE and signal distortion. In almost all DSM research the main focus
has been base station applications. The main differences between base station
and Radio link PA’s is that base stations have a higher output power and lower
envelope bandwidth. This project will investigate if the higher modulated
envelope bandwidth can be accomplished for the relatively low output power in
radio links.
Amplitude
0
1024-QAM
Relative magnitude
-20
-40
-60
-80
-
-
0
0.5
1
1.5
2
2.5
3
3.
Frequency
4
x
8
Figure 4. QAM-1024 envelope bandwidth.
An ideal DPD produces a signal that is the inverse of the system it is meant to
linearize. From a frequency point of view this means an identical spectrum. This
could pose a problem if the bandwidth in the DAC (Digital to analog converter) is
limited, because frequency components outside the sampling range of the DAC
causes aliasing. In this study the DAC bandwidth has been set to 500 MHz, to
avoid aliasing the signal is filtered with an anti aliasing filter that has a cut off
frequency of 170MHz. Since spectral broadening of the RF signal is expected for
DSM this limit will be studied carefully.
3
Characterization of Power Amplifier
In order to calculate the PAE improvement potential a model of the 7 GHz PA
was simulated in ADS. The transfer characteristics were retrieved by sweeping
Gate/Drain Voltage and CW (Continuous Wave) input Power, while measuring
output power and DC power consumption.
The simulations were done using harmonic balance simulation with an RF
frequency of 7 GHz including 5´th order harmonics. In order to accurately predict
the performance with some head room for a 1024-QAM signal Pin was swept
between -40 dBm to 41 dBm and Vdd respectively between 1V and 12V. In order
to prevent input mismatches from affecting the measurements an ideal isolator
was placed at the input port of the PA.
For newer radio generations it is expected that the PA will be run with a lower
conduction angle. Therefore the drain bias sweep was done at two different
quiescent drain currents. The PA was designed for a gate voltage bias of -0.685V
giving a quiescent drain current of to 400mA or 15.5% of Idss. By reducing the
gate voltage to -0.810V Idquiescent was lowered to 200mA which is 6.7% of Idss.
The PA transfer characteristics were retrieved from harmonic balance simulations
in ADS. The result of these simulations with the drain bias voltage swept between
1 V and 10 V can be seen in Figure 5 and Figure 6. It can here be seen that for
relatively high input powers the dominant phenomena is self biasing resulting in
very similar behavior for both bias points. We can in the figures also see a very
sharp decline in gain when the contours go from horizontal to vertical.
Pout[dBm], Idquiescent = 6.7% of Idss
Pout[dBm], Idquiescent = 15.5% of Idss
34
38
40
30
35
26
30
32
34
30
30
28
26
26
24
14
0 8
1
10
3
5
61
5
V
dd
7
8
30
28
26
24
22
22
20
18
16
14
20
18
16
14
12
4
15
30
28
26
24
10 1
6
20
18
16
14
12
2
20
18
18
5
22
20
18
16
14
34
32
24
22
10 16
36
20
28
15
Pin [dBm]
30
20
Pin [dBm]
32
20
24
22
34
25
30
32
36
24
22
25
38
28
28
26
35
36
40
9
[V]
Figure 5. Pout as a function of Vdd and Pin simulated
with Idquiescent = 15.5% of Idss.
10
10
0 8
1
14
12
10
2
3
4
510
6
7
12
8
9
Vdd [V]
Figure 6. Pout as a function of Vdd and Pin simulated
with Idquiescent = 6.7% of Idss.
In Figure 7 and Figure 8 the PAE contour plots for both bias points can be found.
From these graphs we see that the PAE behavior is similar when both drain
voltage and input power is high. For both lower drain voltages and lower input
powers the PAE is increased in the deeper PAE case. Again the self biasing
effect is seen from the fact that the top PAE region and above is very similar for
both figures.
10
PAE, Idquiescent = 6.7% of Idss
PAE, Idquiescent = 15.5% of Idss
40
40
353020 1050
15
in
55
40
50
35
30 25
20
15
45
40
15
25
20
10 15
5
45
40
10
35
30 25
20
15
15
35
30
25
5 20
15
10
0
1
5
1
2
3
4
5
V
6
dd
7
8
9
10
40
40
45
35
30
25
20
15
40
45
10
5
45
55
55
50
55
50
40
10
5
55
45
50
0
20 51105
30
40
50
40
25
45
50
325520
45
4
50
20
0
30
55
25
511002
50
0355
32
40
50
55
45 50
P [dBm]
45
40
45
25
30
25
500
12510
3
5
30
5100
1520
25
30
35
35
Pin [dBm]
35
50
10
15
20
30
35
35
30
25
20
15
10
10
5
5
2
3
4
5
6
7
8
9
10
Vdd [V]
[V]
Figure 8. PAE as a function of Vdd and Pin
simulated with Idquiescent = 6.7% of Idss.
Figure 7. PAE as a function of Vdd and Pin
simulated with Idquiescent = 15.5% of Idss.
In Figure 9 and Figure 10 the transfer characteristics with the gate swept
between -1.5 V and -0.5 V are seen. We can see that for high power levels the
gain is dependent upon gate voltage, with about 4 dB/V, while for small signals
the transistor shuts down under -1V. For a swept input power the maximum PAE
is found to always be over -1V which is before the transistor shuts down.
Pout contours Vdd 10 [V]
PAE contours Vdd 10V
40 18
35
36
32
-0.7
-0.6
18
16
5
12
10-0.5
0 30
-1.5
Gate voltage [V]
Figure 9. Pout as a function of Vgs and Pin simulated
with Vdd = 10V.
24
22
18
16
14
12
10
20
18
16
14
4 68 12
10
20
2480
23
-1
22
26
24
22
6
88
-1.1
24
25 20
22
11
86
14
64
16
1012412
1
20
20
16
08128
24 20 16180 2 18
8 48 46 16
328
14
15 0 26 22
62
416102
4
16 2
14 4
10
6
-1.2
20
18
16
14
12
10
-0.9
-0.8
28
26
26
26
10
22
2 4
0
-1.3
30
24
20
18
16
14
1
102
68
-1.4
34
32
30
28
26
22
5
0
-1.5
34
30
24
18
2022
24
12
10
2 4 42
28
26
34
8
25
20
20 18
16
14
12
10
8
15 6 4
2
0
10
38
36
32
24
26
30
68
38
36
8
30
16 18
20
22
24
10
Input power [dBm]
38
Input power [dBm]
35
24
18
1618
1420
16
22
6
40
8
6
-1
4
2
-0.5
Gate voltage [V]
Figure 10. PAE as a function of Vgs and Pin.
Recalling that the average input signal power is below 20 dBm wee see that the
potential PAE improvement is very small.
3.1
Calculation of supply modulation signals
Knowing the desired output signal, we can calculate the set of supply bias and
input signal that produces not only the correct output but also maximizes the
PAE. From the PA transfer characterization constant contour plots were
calculated using MATAB resulting in Figure 11 and Figure 13. Numerically the
point on each constant output power contour in Figure 11 that maximizes PAE
was chosen, resulting in the black curve seen in both figures. Since this curve is
not monotone at some points and also irregular it would make a poor mapping
function. For that reason a polynomial fit of this black curve was used to calculate
the full bandwidth control signals. To constrain the mapping functions
appropriately a minimum gain and a minimum drain voltage was set. The
minimum gain requirement was set to 5 dB since this is comparable with the
performance of the DPD employed today. Examples of such a mapping function
can be seen in Figure 12. As described in the next section a DPD is implemented
using a 2DLUT, the DPD calculated Pin that corresponds to each output value
via the Vdd polynomial is shown in Figure 14.
PAE contours with optimum PAE drain trace
Pout contours with optimum PAE drain trace
40
40
35
30
36
30
Input power [dBm]
Input power [dBm]
30
34
36
25
28
34
32
20
15
10
24
0
1
2
30
28
26
24
22
20
18
22
20
18
18
5
32
30
28
26
26
22
20
16
14
12
10
3
16
14
12
4
5
6
7
8
1020
0165 35
e- 0
4045
8 27 5
25 5. 79
2
5 30
55
20
50
40
15
15
0
50
45
40
35 30
25 20
15
15
10
10
5
5
1
2
3
4
5
6
7
8
9
10
Drain voltage [V]
Figure 11. Optimum PAE trajectory and polynomial
fit on top of with constant P0out contours.
Figure 13. Optimum PAE trajectory and polynomial
fit on top of constant PAE contours.
Input power trajectory for optimum PAE
Drain voltage polynomial for optimum PAE
40
Max PAE trace
Polyfitted max PAE trace
35
Pin Max PAE trace
Pin Vdd polynomial trace
12
10
30
Input power [dBm]
Drain voltage [V]
45
40
35 30
25
20
5
10
5
50
55
35 30
25 20
10
16
14
12
10
9
10
-006151020
5.795827e
25
35
40
30
45
50
55
45
Drain voltage [V]
8
6
4
25
20
15
10
5
2
0
0
-006
827etrace
5.79
Max
PAE
5
15 1020
Polyfitted 30
max
25 PAE trace
35
40
45
50
38
32
24
Max PAE trace
Polyfitted max PAE trace
35
0
5
10
15
20
25
30
35
40
Output power [dBm]
Figure 12. Example of drain voltage polynomial
fitted onto the optimum trajectory.
0
5
10
15
20
25
30
35
40
Output power [dBm]
Figure 14. Example of Pin trajectory calculated
from the Vdd polynomial.
3.2
Input signal calculation
In Ericsson MINI LINKS system the PA phase and amplitude distortion is
compensated for by a polynomial DPD. These polynomials are calculated as the
inverse of the amplitude and phase shift as a function of output power, commonly
referred to as AM/AM and AM/PM. This polynomial predistortion could
theoretically be employed also for a supply modulating system given that the
memory effects are minor.
In this report special attention has been given to bandwidth reduction. Bandwidth
reduction means there no longer exists a 1:1 relation between signal amplitude
and supply signal. The input signal now has to be calculated using either a
2DLUT or a 2 dimensional polynomial. The implementation chosen for this study
consisted in 2 2DLUTs (2 dimensional look up table) which represent the inverse
of the PA and therefore through interpolation delivers the input signal that
cancels the nonlinearities of the modulated PA.
Calculating this inverse 2DLUT was done using the data produced by the PA
characterization. The sets of Pin and Vdd that produces constant Pout seen in
Figure 15 were interpolated onto a mesh grid comprised of the desired Vdd and
Pout range. This resulted in the 2DLUT seen in Figure 16 which is defined for all
points below unit gain.
Constant Pout contours
35
40
20
Input Power [dBm]
30
30
25
10 1
5
20
10
10
0
0
-5
-20
-20
2
4
6
8
Drain Voltage [V]
10
5
0
0
15
5
0
-5
-5
-10
-10
-15
-20
-25
-30
-30
-15
-20
-25
-30
-35
-35
-35
-30
-25
-25
5
-10
-20
-20
-25
-40 -3 0
-15
-15
10
-15
35
25
20
10
-5
-2 5
-20
3530
25
20
15
0
-10 -2 0
-10
-10
-15
-30
0
-5
-5
-10
5
0
5
5
5
0
10
20
20
15
15
10
5
10
25
25
25
20
15
35
30
5
20 22105
Pout [dBm]
30
25
35
35
20
-10
3035
30
30
0
2DLUT Pin (drain adjusted amplitude DPD)
40
40
12
Figure 15. Constant Pout contours produced using
the data from the PA characterization.
2
4
6
8
Drain Voltage [V]
10
12
Figure 16. 2DLUT PAE as a function of Vdd and Pin
simulated with Idquiescent = 6.7% of Idss.
What is important to consider when calculating the inverse 2DLUT is where the
amplifier is intended to operate. Supply modulation means in the end to drive the
PA in compression over a large portion of the amplitude range. This means that
accuracy in the area close to the non defined zone is of great importance. By up
sampling the mesh grid and also the Pout contour density for high output powers
this issue can be resolved.
3.3
Analysis of PAE improvement potential
The PAE improvement potential has been evaluated numerically in MATLAB
using the model created by the PA characterization. This was done by employing
the following process:
1 Compute random 1024-QAM IQ signal.
2 Control RF bandwidth using raised cosine filter with a roll off factor of 0.17.
3 Normalize to desired output signal power.
4 Map output power to supply signal.
5 Through 2DLUT interpolate to find input signal.
6 Using a LUT model of the PA the DC power consumption was calculated.
Based on the PA characterization the possible PAE performance was estimated
for the three modulation schemes. In the sections that follow the theoretical
average PAE, for various signal powers is calculated. All evaluations in this
section are done with a 1024-QAM signal with PAR limited to 9 dB. These
calculations were done in MATLAB and without taking envelope amplifier
efficiency into account.
3.3.1
Dynamic drain bias modulation PAE improvement potential
Using data from the PA characterization the PAE as a function of output power
for various drain voltage levels is plotted in Figure 17. Note that the gate bias is
set to 15.5% of Idss. The dynamic range over which the PAE can be increased by
controlling the drain voltage is compared with the probability distribution of a
1024-QAM signal. We see that the probability distribution is well covered within
the Vdd range between 1V and 10V. From this information we can infer that
further reducing the gate controlled Idquiescent will not substantially affect the
maximum achievable PAE.-
Figure 17. PAE as a function of output power for various drain voltage supply levels. Comparing the dynamic
range for drain modulation with the probability distribution of a 1024-QAM signal. Vdd range from 1V to 12V,
with gate bias set to 15.5% of Idss.
From the optimum PAE trajectory a polynomial fit was calculated. This
polynomial was then applied to the RF signal where after the PA characterization
was used to predistort the input signal and estimate the PAE. The result of these
simulations can be found in Table 1. These PAE numbers refer to the PA PAE
not taking the envelope amplifier into account.
Table 1. Simulated maximum PAE for dynamic drain bias modulation.
Average Pout
Average PAE
30dBm
53.6%
28dBm
51.1%
26dBm
47.1%
3.3.2
Dynamic gate bias modulation PAE improvement potential
In Figure 18 the dynamic range of gate modulation can be found. This shows the
same message as seen earlier that the maximum average and peak PAE is
much lower than for the drain modulation. But apart from that it becomes evident
that the improvement potential, which is showed as the area between the blue
and black line is very minor.
‘
Figure 18. PAE as a function of output power for various gate voltage supply levels. Comparing the dynamic
range of gate modulation with the probability distribution of a 1024-QAM signal. Vgs from -1.5V to -0.5V.
The PAE simulation showed that the maximum average PAE using gate
modulation on this 7 GHz PA for a 1024 QAM signal is 18%.
3.3.3
Switched drain bias modulation PAE improvement potential
In instead of dynamic modulation the drain supply it was switched between two or more discrete levels. In
Figure 19 the PAE potential of such an implementation is found. The drain
voltage levels are computed to within 0.25V be the PAE wise optimum set of
levels. Since it is required to reach the highest peak one of the levels must
always remain at 10V. From this simulation we see that an improvement to a
PAE between 25% and 30% is possible given that it is possible to liberalize the
distortion caused by switching.
Figure 19. PAE improvement potential estimation for switched drain modulation. The term switch margin
refers to the time between when a switched is set on and the value is reached.
The evaluation of the three different modulation methods led to the conclusion
that the dynamic drain modulation was going the be the main focus. Specifically
the bandwidth reduction schemes and their implications is the focus for the
remainder of the study.
4
Peak tracking bandwidth reduction
For transistors in general, the dissipated power is increased with the operating
frequency. Envelope amplifiers specifically follow the same relationship showing
efficiency degradation and also linearity issues for higher signal bandwidths. This
is one of the limiting factors for supply modulation [1]. It is with this knowledge in
mind the idea of reducing the supply modulation signal bandwidth arises. Filtering
however is not as straight forward as simply applying a FIR LPF, since the top
supply voltage is needed in order to maintain gain and linearity for the peaks.
This leads to the conclusion that a PTLPF (Peak Tracking Low Pass Filter) is
needed.
In [1] the iterative filter described by Figure 20 was found. The principle of
operation is to add the positive difference between the original polynomially
calculated Vdd signal and its low pass filtered version. For every iteration the
peaks will approach the desired level and thus produce the PAE optimized supply
signal for a given cutoff frequency.
Figure 20. Schematic description of the iterative peak tracking low pass filter.
The design of this filter has the interesting characteristic stemming from the super
positioning of the filtered signals. If the slope of the LPF applied to the residual
signal is not steep enough the stop band content will add up. Letting these high
frequency components propagate though to the EA would most likely cause
linearity issues related to mismatches between supply and RF signal. During this
master thesis project this problem has been solved by implementing the LPFs
with 10´th order elliptical FIR filter which ha a very steep cutoff frequency and a
very non linear phase response. The phase linearity issues were not addresses
but rather ignored by using the “filtfilt” function in MATLAB, which lets the signal
run through the filter in both forward and reverse direction resulting in a constant
phase response.
Recalling from the conceptual description we could note that time alignment
between supply and RF signals is crucial for the linearity and to a certain extent
also efficiency. This is of course also an important consideration in the filter
design, since frequency dependent delay would cause poor synchronization
which in worst case means undesired compression and nonlinearities.
The peak tracking property is necessary while the level of the lower areas is a
trade off between an increase in PAE and reduced gain and linearity.
4.1.1
PAE vs EA bandwidth trade off
Employing the peak tracking low pass filtering algorithm described in the method
section resulted in the data presented in Figure 21. The result is a trade off
number of filter iterations and the low pas filter cut off frequency. The limiting
factor for number of filter iterations is the time delay it is translated into when
implementing the filter. Low pass filter is in turn traded against efficiency in the
EA achievable
PAE-trade off analog drain modulation
43 .94 66
52
41 .11
39
66
.94
43
Filter iterations
41 .1152
8
38 .2
5
35 .45 2
2
32 .62 1
85
38 .2
8 39
29
.78 32 .62 35 .45 25
12
98
18
2
.46 21 .29 4 .12 7 26 .95
58
85
1
44
10
5
0
5
26 .9
71
24 .12
15
8
21 .29 5
18 .46 44
20
8
29 .78 9
25
35 .45 25
30
26 .95 85
35
24 .12 71
21 .29 58
40
32 .6212
29.7898
45
38.2839
50
0
5
10
15
20
25
30
41 .11
52
38 .28 3
9
3
32 .62 5 .45 25
12
35
40
45
50
Low pass filter cut off frequency [MHz]
Figure 21. Trade off between LPF cut off frequency and number of filter iterations for bandwidth reduced
dynamic drain modulation.
From Figure 21 we can conclude that this bandwidth reduction scheme requires
more than 30 filter iterations. Further for filter cut off frequencies over 25 MHz the
PAE dependence is not as strong.
5
Linearity evaluation
5.1
DPD performance
The DPD performance is evaluated in terms of the ability to adjust for the
distortion in the PA. A common way to view the distortion is by looking at the
AM/AM and AM/PM behavior. For a few selected envelope bandwidths this can
be seen in Figure 22 and Figure 23. As can be seen in the figures, the static and
infinite bandwidth lines constitute mapping functions between output power and
control signals. But for the bandwidth reduced cases they turn into more of a
statistical cloud. Already the non linear and non monotonic behavior of the infinite
bandwidth curves would be a challenge for the DPD. This also leads to the
conclusion that maybe there is a different way to calculate the optimum trajectory
that minimizes that gain and phase variations.
For the AM/AM case the infinite and static lines set the boundaries for the
probability clouds of the bandwidth reduced ones. The relation between the lines
and clouds is slightly different for the AM/PM the case, instead the clouds
become more similar to the static line for low frequencies and vice versa for high
frequencies.
Figure 22. Predicted AM/AM behaviour when
empoyeding DSM.
Figure 23. Predicted AM/PM behaviour when
employing DSM.
One way of describing the DPD performance is to look at the RF frequency
spectrum. In Figure 24 the RF frequency response with the DPD turned off is
seen. This can be compared with Figure 25 where the DPD is deployed.
Recalling that the DPD is an inverse 2DLUT of the PA characterization this result
is to be expected verifying that the DPD in fact is an accurate inverse.
Simulated output spectrum Pout30[dBm] without DPD
Simulated output spectrum Pout 30[dBm]
20
20
MASK
Reference
Vdd static
Vdd BW 1MHz
Vdd BW 5MHz
Vdd BW 20MHz
0
-10
-20
-30
-40
-50
-60
-70
-80
MASK
Reference
Vdd static
Vdd BW 1MHz
Vdd BW 5MHz
Vdd BW 20MHz
10
Relative magnitude [dB]
Relative magnitude [dB]
10
0
-10
-20
-30
-40
-50
-60
-70
-3
-2
-1
0
1
Frequency [Hz]
2
-80
3
-3
-2
-1
0
1
Frequency [Hz]
8
x 10
Figure 24. MATLAB simulation comparing desired
output signal with simulated response for an
output power of 30dBm.
2
3
8
x 10
Figure 25. MATLAB simulation comparing desired
output signal with simulated predistorted response
for an output power of 30dBm.
Another way of evaluating DPD performance is through the MSE(Mean Squared
Errror). In Table 2 the MSE for output powers of 28dBm and 30dBm at different
level of supply modulation is found. In this table as in the spectrum pictures we
can clearly see that the DPD is accurate for static drain voltages and then
gradually degrading for increasing modulation bandwidth. For drain voltage
modulation above the RF signal envelope bandwidth the EVM will improve.
Table 2. MSE for different output power and modulation bandwidth.
Vdd
static
Vdd BW
1MHz
Vdd BW 5
MHz
Vdd BW 20
MHz
Pout = 28dBm
-82.1 dB
-68.3 dB
-54.2 dB
-45.7 dB
Pout = 30dBm
-70.3 dB
-68.7 dB
-60.3 dB
-48.5 dB
5.2
Input spectrum
In Figure 26 and Figure 27 the input spectrum to the DAC (Digital to analog
converter) is seen. As expected the spectrum is similar to that of the output
spectrum without DPD in Figure 24.
The input spectrum does not have to conform with the spectrum mask as the
output does. Instead the input spectrum is not allowed to spread further than the
DAC converter bandwidth. For a DAC bandwidth of 500MHz the spectrum is
required to be lower than -55dBc at 170MHz away from the center frequency. If
the DAC instead has 800MHz of bandwidth this limit is shifted to lower than 55dBc at 250 MHz away from the center frequency. The 4 vertical black lines are
positioned at 170MHz and 250 MHz away from the center frequency. From the
figures we see that the spectrum width requirement for the 500 MHz DAC is not
satisfied for any other than the static drain voltage. On the other hand they all fit
nicely inside the 800 MHz DAC spectral requirements.
Input Spectrum for 28[dBm] average Pout
Input Spectrum for 30[dBm] average Pout
20
20
MASK
Reference
Vdd static
Vdd BW 1MHz
Vdd BW 5MHz
Vdd BW 20MHz
0
-20
0
-20
-40
-40
-60
-60
-80
-3
-2
-1
0
1
2
3
8
Frequency [Hz]
x 10
Figure 26. Spectral growth of input spectrum with
DSM for 30dBm average Pout.
MASK
Reference
Vdd static
Vdd BW 1MHz
Vdd BW 5MHz
Vdd BW 20MHz
-80
-3
-2
-1
0
1
2
3
8
Frequency [Hz]
x 10
Figure 27. Spectral growth of input spectrum with
DSM for 28dBm average Pout.
6
Verification of PAE potential calculations
The PAE improvement potential evaluation algorithm implemented in MATLAB
has to be verified by simulating a PA controlled by the calculated signals. This
was done using the MALAB co simulation package ADS Ptolemy. The PA was
fed with a modulated drain voltage and a predistorted RF carrier signal voltage.
Both of these signals were computed in MATLAB and via ADS Ptolemy fed to
PA. During the simulation the RF output and DC power consumption was
measured to compare linearity and DC power consumption.
6.1
Verification of PAE improvement calculations
By simulating the PA model in ads, with the MATLAB calculated signals the hope
was to evaluate the validity of the MATLAB calculations. This was unfortunately
unsuccessful since the simulations turned out as noise as can be seen in the
figure below. What could be evaluated though was the PAE estimation which if
compared with the MATLAB calculation in Figure 21 can be seen to match up to
simulated values of Table 3.
Table 3. PAE improvement verification
Vdd – BW
PAE
DPD-state
Average Pout
Figure 28
0 MHz
14.8 %
ON
30 dBm
Figure 30
0 MHz
14.9 %
OFF
30.04 dBm
Figure 29
20 MHz
37 %
ON
30.05 dBm
Figure 31
20 MHz
38 %
OFF
30.22 dBm
15
15
14
14
13
13
12
12
11
GaindB
GaindB
In the figures below the stochastic behavior for higher output powers can be seen
where the PA reaches compression early. We can note that the gain at all times
is kept above 9 dB when the DPD is turned of. This tells us that the DPD range is
not a limiting factor.
10
9
11
10
9
8
8
7
7
6
6
5
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
P_out_dBm
Figure 28. ADS verification, VDD static DPD ON.
5
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
P_out_dBm
Figure 29. ADS verification, Vdd 20 MHz DP ON.
15
14
14
13
13
12
12
11
10
9
GaindB
GaindB
15
11
10
9
8
8
7
7
6
5
-2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38
P_out_dBm
Figure 30. ADS verification, Vdd static NO DPD.
6
5
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
P_out_dBm
Figure 31. ADS verification, Vdd 20 MHz No DPD.
7
Complete PA chain PAE
Pre-amp
Driver
PA
Figure 32. Transmitter amplifier stages.
The complete PA chain in this study consists of 3 consecutive stages in series.
These are seen in Figure 32 and are the preamplifier, driver and PA. The ADS
model that was characterized is of the last stage which is the PA. Since models
for the proceeding stages where not available they were approximated by the PA
model with downscaled bias voltages. The preceding stages have approximately
the same gain but are backed off from the 1dB compression further point to
improve linearity. Non linearity in early stages would otherwise be amplified in the
later ones. The drawback of this is poor efficiency, but since their output power is
much lower the system loss is less affected.
Table 4. Amplifier back off relative to the PA of the 3 steps in the amplifier chain.
Relative amplifier back off
Pre amplifier
Driver
PA
10 dB
3 dB
0 dB
Modulating the drain voltage of the PA implies operating it closer to the point of
compression for a large part of the Pin,RF range. When the PA goes into
compression gain is lost, this means more power has to be delivered by the
linear and inefficient preceding stages.
By approximating the DC characteristics and transfer function as a scaled version
of the final stage PA, the DC power consumed in of the preceding stages could
be calculated. This was done using the relative back off of each amplifier as
follows:
(ηEA denotes envelope amplifier efficiency)
Pin. relative(dB)
= Pin,PA(dB) - relative amplifier back off(dB)
PAE
= PAE2DLUT, estimator{ Pin. relative(dB) }
Pout. relative(dB)
= Pout, 2DLUT, estimator{ Pin. relative(dB) }
PDC, estimated. (W)
= (Pout. relative(dB) - Pin. relative(dB) ) / PAE / ηEA
In Table 5 the complete PA chain PAE with different EA efficiencies and different
number of PA stages modulated is seen. The modulation evaluated here is
10MHz dynamic drain modulation.
By introducing supply modulation on the last PA stage only the individual stage
PAE increases by 13 percentage units but the total PAE is still only increased by
4.5%. But, if we introduce modulation also on the driver stage an additional 7.5%
is increased in the total PA PAE.
Also recall here that modulating the supply of the Driver and Pre-driver will
introduce additional non linearity and loss of gain.
Table 5. PAE for compete and individual gain stages in PA, with and without drain modulation. Vdd bandwidth
10 MHz.
EA efficiency
Total PAE
Individual PAE
(not considering
EA efficiency)
Number of stages with modulated supply
0/3
1/3
2/3
3/3
Pre-driver
1.3%
1.4%
1.7%
18.2%
Driver
2.9%
3.2%
22.3%
22.3%
PA
15.6%
28.5%
28.5%
28.5%
10.6%
13.3%
18.6%
20.3%
10.6%
14.2%
20.6%
22.8%
10.6%
15.1%
22.64%
25.32%
80%
90%
100%
8
Envelope amplifier
A conceptual envelope amplifier design was chosen after studying literature and
consulting the Power department at Ericsson. Apart from efficiency, good linearity
over the desired bandwidth is important in order to provide the PA with the
desired modulated supply voltage. The chosen design that provides both high
efficiency and high bandwidth, is comprised of a low speed high efficiency
switching stage and a high speed linear amplifier in parallel. See Figure 33.
Figure 33. Conceptual description of the Envelope amplifier.
The switching stage is usually a very efficient buck converter switching stage that
is followed by a large inductor. This inductor together with the load resistance
serves as a LPF (Low Pass Filter) blocking high frequency distortion generated in
the switch. From the collaboration with Ericsson power it was found that their
implementations used 16 μH series inductors. For this simulation the PA was
modeled as a 10 ohm resistor. This 10 Ohm resistor in series with the 16uH
Inductor yields a cut off frequency ~100 KHz. This is too low to have any impact
from a DSM point of view.
In the static bias case the drain voltage is provided with a similar kind of DCDC
converter and since the efficiency of those is comparable the switch stage
efficiency will not be investigated or accounted for in this study.
8.1
Implementation
The linearity and bandwidth performance is limited by the linear amplifier, while
efficiency requires both stages to have low power dissipation. The EA design
presented by Ericsson power was designed for base station applications, this
means lower bandwidth and higher power compared to the MIN ILINK EA in this
study.
The linear amplifier has to accurately track the desired drain voltage, operating
like an ideal variable voltage source. The PA drain current is the purely set by the
drain source impedance controlled by the RF input signal and bias. Since the
linear stage output voltage must be able to swing both up and down a push pull
configuration inside an Op-amp feedback loop was chosen as basic design. The
push pull stage consists of a pair of complementary bipolar transistors, which has
the best performance for high frequency, high current applications. See Figure 35
for a description of the feed back loop.
Figure 34. Basic push pull feedback loop.
The base emitter voltage drop in a BJT causes a “knee” in the transfer function of
the amplifier in Figure 35. This has to be compensated for in order to get a linear
behavior. By placing two diode coupled transistors between the bases of the two
transistor this is taken care of, but only for high output currents. In order to
provide a constant voltage drop that is independent of the op-amp output a
current mirror stage is placed as can be seen in Figure 35. In this figure we also
see two current buffers on the base of both BJTs. A current buffer is a unit
voltage gain amplifier boosting the current. The reason for having such buffers is
that better frequency performance in the BJTs can be achieved, if devices with
lower gain are chosen.
Figure 35. Envelope amplifier conceptual layout.
There are 4 sets of resistors in this circuit; these are the R_mirror, R_early,
R_stab, R_feedback. R_mirror has the functionality of setting the current mirror
current together with the normally much lower R_early which has the function of
reducing VCE dependence on the current in the current mirror. R_stab was found
to solve an initial problem of oscillation between the PNP, NPN transistors.
Finally we have the R_feedback which is the most sensitive design parameter of
the linear amplifier as it sets the feedback speed and thereby cutoff frequency.
Setting it too low though will result in an unstable circuit.
The PA model provided has a high complexity and was therefore initially modeled
by a resistor. From the MATLAB simulations it was found that the DC input
impedance at the PA drain port has an average of 10 ohm for a 10MHz
modulated drain signal. A 10 ohm resistor was therefore selected as a simplified
PA model. For final testing off course this resistor was replaced by the complete
PA ads model.
8.2
Choosing components
The optimization of the linear amplifier is a trade off between, linearity, energy
consumption, bandwidth, voltage swing and current swing. The major challenge
lies in achieving a high bandwidth without causing instability, distortion and major
power dissipation. Voltage swing is determined by the DC supply level, this sets
the minimum DC supply level and apart from some design head room a higher
level is rarely desirable, since it will increase the dissipated power. In turn the
current swing is set by Icmax in the BJT´s, it is therefore desired to choose a
sufficiently large transistor. But for BJT´s there exists an approximate inverse
relation between ft, fmax and Icmax. This implies that a too large transistor size
degrades the frequency performance. In addition to this the current gain of the
BJT´s has to trade against output current from the OP-amp and current buffer.
8.2.1
Components
The Operational amplifier in the circuit is from Texas instruments and the model
number is THS3001 which has a 3dB bandwidth of 420 MHz and a maximum
output current of 175 mA. This meets the requirements of the EA with head room.
A bandwidth wise critical part of the circuit was the current buffers. The
performance wise best buffer found was the HA-5002 from Intersil. This Buffer
has a bandwidth of 110MHz and an output power of ±200mA.
The most sensitive part of the linear amplifier is the output stage NPN PNP
transistors. This part has to be chosen together for maximum predictability. The
two components chosen were model NJT4030p and NJT4031n from ON
semiconductor. They featured a maximum collector current of 5 A and an fT of
160 MHz and 215 MHz respectively.
The knee compensating diode coupled NPN transistors have the model number
2n3904 and is available from STMicroelectronics. The specification for the
transistor is mainly to have a high fT which for this model was 250 MHz.
In the current mirror two identical transistors are placed that need only have low
power dissipation and stable operation. These transistors were chosen from
STMicroelectronics and the model BC547C.
8.3
Simulations
Using ADS, a simulation model of Figure 35 was designed and optimized. In
order to cut simulation time while designing a few simplifications were made. The
switched BUCK stage was in the simulation replaced with a voltage source
followed by an RF choke.
As mentioned before the PA was initially simulated with an ideal resistor. The
value of this resistor was selected from the average value of the drain source
impedance with a 10 MHz dynamic supply modulation.
8.4
EA simulation results
The EA design procedure consisted in several steps involving various ADS
simulation engines. The final design though was evaluated using transient
simulation in order to make sure the design was stable for all frequencies.
InTable 6 a display of the complete PAE of the final stage PA and EA combined
is shown. This shows how the increased power dissipation in the EA for higher
frequency still is outweighed by the increased PAE of the PA.
Table 6. EA power dissipation for increasing bandwidth.
Vdd bandwidth
1 MHz
5 MHz
10 MHz
20 MHz
30 MHz
DC power consumed by PA W
4,8
3,4
2,7
2,6
2,26
DC power dissipated in Linear EA W
1,2
1,7
1,9
1,9
2,1
Pin W
0,1
0,1
0,1
0,1
0,1
Total dissipated power W
6,1
5,1
4,6
4,5
4,4
Pout W
1,0
1,0
1,0
1,0
1,0
PAE PA
19,0%
24,0%
31,0%
38,0%
40,0%
PAE EA+PA
15,2%
18,0%
19,6%
19,9%
20,4%
PAE reference without DSM
15,6%
40 MHz
2,0567
2,1
0,1
4,2
1,0
43,0%
21,2%
In addition to trading the linear EA efficiency performance against PA PAE the
EA linearity is of great concern. In Figure 36 the MSE as a function of frequency
swept between 1 and 101 MHz is seen. What can be seen here is that the
performance starts to degrade severely after around 40 MHz.
20
10
MeanSquareError
0
-10
-20
-30
-40
-50
0
10
20
30
40
50
60
70
80
90
100
110
frequency
Figure 36. Linear Envelope amplifier MSE as a function of frequency.
The linear amplifier has a predictable and desired response for frequencies up to
40 MHz. An example of the time domain response can be seen in Figure 37.
Comparing this figure with a simulation where the resistor is replaced with the
real PA model yields the result seen in Figure 38. The high frequency noise
stems from the RF envelope frequency controlling the drain current. The drain
current variations that are outside the bandwidth of the EA will cause the drain
voltage to deviate from the desired causing the RF signal to be distorted.
11
10
Vdd_out, V
Vdd_in, V
9
8
7
6
5
4
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
time, usec
Figure 37. Comparison of input and output signal of the linear amplifier for a
QAM-1024 modulated drain voltage signal bandwidth reduced to 5 GHz
simulated with a resistive load.
11
10
Vdd_out, V
Vdd_in, V
9
8
7
6
5
4
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
time, usec
Figure 38. Comparison of input and output signal of the linear amplifier for a QAM1024 modulated drain voltage signal bandwidth reduced to 5 GHz simulated with
an ADS model PA as load.
An attempt at resolving this issue was made by placing a LPF between the EA
and PA and thereby isolating them for high frequencies. The idea was to lower
the drain signal bandwidth down to around 10 MHz and place the cut off
frequency of the LPF below the bandwidth of the linear amplifier at around 20-30
MHz. By having this overlap between the EA bandwidth and the LPF it was
assumed that good isolation could be achieved. Unfortunately these attempts
were unsuccessful and this seems to be an area where continued investigations
are needed.
9
Conclusions
The PAE improvement achieved in this project shows a potential improvement of
the final stage PA from 15% to 20%. This number takes efficiency of the
inefficient EA into account. Simulations of the entire amplifier chain indicate that
the complete amplifier PAE can be improved from 10% to 20% with an EA
efficiency of 80%.
The polynomial mapping signal envelope to drain voltage should not only be
optimized for maximum PAE. By choosing the polynomial carefully large gains in
linearity can be achieved.
From the PA characterization it was found that the potential PAE improvement
with gate modulation is much lower than with drain modulation.
Driving the amplifier in compression over a wider range of signal amplitudes
causes spectral growth. The heavily distorted signal poses difficulties not only for
the DPD, but also for the DAC. Spectral growth in the non predistorted signal is
when applying predistortion translated into spectral growth of the signal going
into the DAC. Aliasing caused by input signal broadening was for this PA found to
be a problem requiring a wideband DAC.
Bandwidth reduction introduces the possibility of supply modulator bandwidth
lower than the RF signal envelope bandwidth. For such a system termination of
RF envelope induced drain current variations is important and must be taken
care of through a filter between EA and PA.
For radios with such high envelope bandwidths as in this project the main
bottleneck is modulator bandwidth. The discrete components used in the
envelope amplifier in this study limits the maximum operating frequency.
Dynamic supply modulation remains an interesting alternative for enhancing the
efficiency of low power PA´s, though very high signal bandwidths are a hard
obstacle to overcome.
10
Future work
Further investigation into envelope amplifier architectures that has increased
bandwidth and improved efficiency is the key to feasible envelope tracking. For
high frequency radio links operating a lower power levels the main advantage is
the low bias voltage that should enable smaller devices with lower parasitics.
The tradeoff between linearity and PAE for the drain voltage polynomial should
be evaluated further.
The problem of envelope induced current ripple in the drain supply bias for low
bandwidth supply modulators has to be solved in order to implement the
bandwidth reduction scheme.
Since linearity is very critical in amplifier design it is important that a simulation
test bench is developed.
11
References
[1] J. Jeong, D. F. Kimball, M. Kwak, C. Hsia, P. Draxler, and P. M.
Asbeck, “Wideband envelope tracking power amplifiers with reduced bandwidth
power supply waveforms and adaptive digital predistortion techniques,” IEEE
Trans. Microwave Theory Tech., vol. 55, no. 12, pp. 2757–2765, Dec. 2007
[2] F. Wang, D. Kimball, J. Popp, et al., "Wideband envelope elimination and
restoration power amplifier with high efficiency wideband envelope amplifier for
WLAN 802.11g applications," in Proceedings of the Microwave Symposium
Digest, pp. 645-648, La Jolla, Calif, USA, June 2005.
[3] E. Cipriani, P. Colantonio, F. Giannini, R.Giofre, L. Piazzon, ”Effects of
Envelope Tracking Technique on an L-band Power Amplifier” IEEE
Mediterranean Electrotechnical Conference, pp. 348 - 352 , April 2010
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising