J. Stevenson Kenney
September 9, 2002
C S
T
Advanced Architectures for Predistortion Linearization
of RF Power Amplifiers
J. Stevenson Kenney, Youngcheol Park, and Wangmyong WooSchool of Electrical and Computer Engineering
Georgia Institute of Technology
Presented atThe IEEE Topical Workshop on Power Amplifiers
September 9, 2002La Jolla, CA
2
J. Stevenson Kenney
September 9, 2002
C S
T
Acknowledgement
• This work was supported in part by Danam USA, San Jose, CA.
and by
• The Yamacraw Design Center, an economic development project funded by the State of Georgia.
• Thanks also to Sirenza Microdevices, and Ericsson USA for donating the power amplifiers used in this study.
3
J. Stevenson Kenney
September 9, 2002
C S
T
Outline
• Introduction to Predistortion• LUT Updates using Sub-sampling
Receivers• RF Envelope Predistortion• Summary and Conclusions
4
J. Stevenson Kenney
September 9, 2002
C S
T
Predistortion Concept
PA response after predistortion algorithm
PA Response
Input operating range without Pre-D
Input operating range with Pre-D
Average power
Peak power
PAPR
This back-off value is determined to give no
significant nonlinear region for peak input power
This back-off value is (Ideal limiter saturation point-PAPR
of input signal)
5
J. Stevenson Kenney
September 9, 2002
C S
T
Correction Techniques for Cellular Base StationsCorrection Techniques for Cellular Base Stations
Correction Technologies
Correction Capability*
Correction Bandwidth
Relative Cost
Feed Forward 25-35 dB > 100 MHz High
Envelope Feedback 10-20 dB < 5 MHz Med
Analog Pre-Distortion 5-10 dB > 25 MHz Low
Adaptive Pre-D 10-20 dB > 50 MHz Med
Correction Technologies
Correction Capability*
Correction Bandwidth
Relative Cost
Feed Forward 25-35 dB > 100 MHz High
Envelope Feedback 10-20 dB < 5 MHz Med
Analog Pre-Distortion 5-10 dB > 25 MHz Low
Adaptive Pre-D 10-20 dB > 50 MHz Med
* IMD Correction based on 8-Tone Continuous Random Phase
6
J. Stevenson Kenney
September 9, 2002
C S
T
Predistortion Linearization
Depends on DSP computational
capability
GoodModerateDigital Baseband Pre-D
No I/Q streamrequired
ModerateModerateAdaptive Analog Pre-D
(Work function)
Simple Implementation
LowHighOpen Loop Analog Pre-D
CommentsRelative IMD Correction
BandwidthPredistortionTechnology
7
J. Stevenson Kenney
September 9, 2002
C S
T
Baseband Pre-DistortionBaseband Pre-Distortion
• Adaptive Filter to Minimize MSE• ADC/DAC Requirements:
– 20 MHz BW ⇒ 100 MHz with 5th order IMD– 3X over sample ⇒ 300 Msps (on both I and Q)
• DSP Speed: >6⋅108 MAC/sec– 0.18 µm CMOS: ~1.5 nW/MAC– ⇒ ~1W DC power– LUT update must also be performed
• Adaptive Filter to Minimize MSE• ADC/DAC Requirements:
– 20 MHz BW ⇒ 100 MHz with 5th order IMD– 3X over sample ⇒ 300 Msps (on both I and Q)
• DSP Speed: >6⋅108 MAC/sec– 0.18 µm CMOS: ~1.5 nW/MAC– ⇒ ~1W DC power– LUT update must also be performed
DSP Pre-D
PAI/Q Stream Input
Σ
DAC
ADC
RF Output
Error Signal
8
J. Stevenson Kenney
September 9, 2002
C S
T
Adapted from R. H. Walden, Performance Trends for Analog-to-Digital Converters, IEEE Communications Magazine, February 1999, pp. 96 - 101.
Trends in Analog-to-Digital Converter Technology
9
J. Stevenson Kenney
September 9, 2002
C S
T
Pre-D Results with Low Power Amplifier (0.5W Class-AB GaAs HFET)
Without Pre-D With Pre-D
10
J. Stevenson Kenney
September 9, 2002
C S
T
Disadvantages of Baseband Pre-D
• Sampling Requirements• DSP speed (power)• Digital I/Q input stream required• LUT update must be performed in
background
11
J. Stevenson Kenney
September 9, 2002
C S
T
Outline
• Introduction to Predistortion• LUT Updates using Sub-sampling
Receivers• RF Envelope Predistortion• Summary and Conclusions
12
J. Stevenson Kenney
September 9, 2002
C S
T
Sampling in Predistorters
• A predistorter samples signals at the input and output of a power amplifier to identify its AM-AM, AM-PM characteristics.
PA
RFDownconverter
A/D Converter
Sample & Hold Circuit
DSP
RFDownconverter
A/D Converter
Sample & Hold Circuit
PredistorterInput Signal
ffOutput Signal
13
J. Stevenson Kenney
September 9, 2002
C S
T
Input Nyquist Rate
• Input Nyquist rate is whenfIF +BW/2=fs-fIF – BW/2
• Again, this is equivalent to the second figure in terms of aliasing
• Therefore, this is 33% of the output Nyquist Rate, considering 3rd order IMD.
0 fs
0 fs’
fIFfs- fIF
14
J. Stevenson Kenney
September 9, 2002
C S
T
Input Nyquist Rate
• The original signal can be reconstructed from the sampled signal if the signal is sampled so that its spectrum is not overlapped in frequency domain.
∑∞
−∞= −−
=k ss
sss TkTt
TkTtkTxtx
/)(]/)(sin[
)()(π
πwhere os
s
FFT
21
≥=
• Input Nyquist rate is the sampling frequency when the highest frequency of the input signal coincides with the lowest frequency of the image signal.
0 fsfIF fs- fIF
15
J. Stevenson Kenney
September 9, 2002
C S
T
Output Nyquist Rate
• With predistortion systems, the analog-to-digital conversion at the output of a PA is traditionally done with ‘above’ Output-Nyquist rate in order to avoid the aliasing at the output spectrum.
0fsfIFfL fH fs- fIF fH,i
= fL,i
16
J. Stevenson Kenney
September 9, 2002
C S
T
Sub-sampling
0 fs=10fIF=2fL fH=3.84
8
fL,i=6.2
fH,i
fs=10MHz
0 fs=5.01fIF=2
fL
fL,i=1.17
fH,i
fs=5.01MHz(67% overlapped)
fs
fIF=2fL fH
BPF
fs<3.55(More than 133% overlapped)
17
J. Stevenson Kenney
September 9, 2002
C S
T
Sampling Requirements for Nonlinear System Identification
• The goal of the sampling system is the pre-D system is not to reconstruct signals, but to identify the nonlinear distortion.
• Therefore, it is not necessary to do Nyquistrate sampling.
• According to the Generalized Sampling Theorem*, a nonlinear system may be accurately identified if the output signal is sampled at the input Nyquist rate.
*John Tsimbinos, and Kenneth V. Lever, “Sampling Frequency Requirements for Identification and Compensation of Nonlinear Systems,” in Proc. ICASSP, vol. III, 1994, pp. 513-516.
18
J. Stevenson Kenney
September 9, 2002
C S
T
Basic Operation of the Generalized Sampling Theorem
NonlinearMapping
g(⋅) = f--1 (⋅)
Samplertk=kTs
x(t)y(t) y(tk) g(y(tk)) LPF
NonlinearMapping
f- (⋅)
x(t) y(t) = f(x(t))
19
J. Stevenson Kenney
September 9, 2002
C S
T
Sub-Sampling ArchitectureADS Simulation
20
J. Stevenson Kenney
September 9, 2002
C S
T
Sub-Sampling ArchitectureADS Simulations (cont.)
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
mag(CH1)
0.0
0.2
0.4
0.6
0.8
1.0
mag
(CH
2)
0 2 4 6 8 10
Mega_Hertz
-150
-100
-50
0
SC
H1
SC
H2
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
mag(CH1)
0.0
0.1
0.2
0.3
0.4
0.5
mag
(CH
2)
0 2 4 6 8 10
Mega_Hertz
-150
-100
-50
0
SC
H1
SC
H2
Figure 1: Simulated AM-AM Characteristic when fs= 10 MHz (includes 3rd order distortions).
Figure 2: Frequency Spectrum when fs = 10 MHz (includes 3rd order distortions).
Figure 3: Simulated AM-AM Characteristic when fs = 5 MHz (70% of output spectrum is overlapped).
Figure 4: Frequency Spectrum when fs = 5 MHz (70% of output spectrum is overlapped).
Input Signal
Aliased Signal
Input Signal
Aliased Signals
21
J. Stevenson Kenney
September 9, 2002
C S
T
Sub-sampling Architecture Test Bed
IF Filter
Lo
PulseGenerator
LPF
880MHz
S/H Circuit
CH1
HP89410
CH2
ErrPowerAmplifier
PCHP4432
IF Filter
Lo
PulseGenerator
LPF
880MHz
S/H Circuit
CH1
HP89410
CH2
ErrPowerAmplifier
PCPCHP4432HP4432
22
J. Stevenson Kenney
September 9, 2002
C S
T
Measurement Results
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6
Vin [V]
Vo
ut
[V]
0
0.1
0.2
0.3
0.4
0.5
0 0.1 0.2 0.3 0.4 0.5 0.6
Vin [V]V
ou
t [V
]
Figure 6: Measured AM-AM Characteristic of SHF-0189 when fs = 10 MHz (includes 3rd order distortions).
Figure 7: Measured AM-AM Characteristic of SHF-0189 when fs = 5.3 MHz (67% of output spectrum is overlapped).
23
J. Stevenson Kenney
September 9, 2002
C S
T
Indirect Learning Predistortion Architecture
Squares Least of methodby calculated
)( of estimate an is)( and
PA,of functiontransfer nonlinear theis )f( where
)),(()()(
)),(()(:Algorithm Basic
1- ⋅⋅
⋅−=
=
fg
nygnzne
nzfny
24
J. Stevenson Kenney
September 9, 2002
C S
T
Predistortion Results
0
2
4
6
8
10
12
14
-100-50050100150200
Overlapped Spectrum of Output Signal [%]
AC
PR
Impr
ovem
ent
[dB
]
ACPR Improvement Using Subsampling Predistortion Architecture (Negative values in x-axis indicates sampling above Nyquist rate of the output signal)
25
J. Stevenson Kenney
September 9, 2002
C S
T
Predistortion Results (cont.)Sirenza SHF-0589: 2W GaAs HFET PA
Figure: ACPR@885kHz measurements over Pin variations.
26
J. Stevenson Kenney
September 9, 2002
C S
T
Predistortion Results (cont.)Motorola MRF-9180 170W Si LDMOS
27
J. Stevenson Kenney
September 9, 2002
C S
T
Impact of Memory Effects
“Memoryless” PA PA with Strong Memory Effects
28
J. Stevenson Kenney
September 9, 2002
C S
T
Impact of Memory Effects (cont.)
29
J. Stevenson Kenney
September 9, 2002
C S
T
Outline
• Introduction to Predistortion• LUT Updates using Sub-sampling
Receivers• RF Envelope Predistortion• Summary and Conclusions
30
J. Stevenson Kenney
September 9, 2002
C S
T
RF Envelope Predistortion
• Predistortion is performed directly on the modulated RF carrier using a high-speed vector modulator
• VMOD is driven by a look-up table (LUT) that is indexed by the instantaneous input power level
• Kusunoki, et al., demonstrated 6-7 dB ACPR improvement without adaptive feedback*
*S.Kusunoki, et al., “Power Amplifier Module with Digital Adaptive Predistortion for Cellular Phone, 2002 IEEE MTT-S Int. Microwave Symp. Dig., pp. 765-8, Seattle, WA, June 4-6, 2002.
31
J. Stevenson Kenney
September 9, 2002
C S
T
Adaptive RF Envelope Pre-Distortion Test BedAdaptive RF Envelope Pre-Distortion Test Bed
32
J. Stevenson Kenney
September 9, 2002
C S
T
FPGA LUT Implementation(Xilinx Vertex-II)
§ LVDS: Low Voltage Differential Signaling§ UART: Universal Asynchronous Receiver Transmitter§ DCM: Digital Clock Manager
33
J. Stevenson Kenney
September 9, 2002
C S
T
RF Envelope Pre-DistortionADS Simulations
847.6 847.8 848.0 848.2 848.4 848.6 848.8 849.0 849.2 849.4 849.6 849.8 850.0 850.2 850.4 850.6 850.8 851.0 851.2 851.4 851.6 851.8 852.0 852.2 852.4847.4 852.6
-85
-80
-75
-70
-65
-60
-55
-50
-45
-40
-35
-30
-25
-20
-15
-90
-10
Mega_Hertz
(a) 2x Sampling
(b) w/o Pre-D
(c) 4x Sampling
(d) Source
dBm
-20
-10
0
10
20
30
6 7 8 9 10 11 12 13 14
nBits
AC
PR
Impr
ovem
ent,
dB
ADC for DSPADC for LUTDAC for VMOD
34
J. Stevenson Kenney
September 9, 2002
C S
T
RF Envelope Pre-D Test Bed
LUTτ
35
J. Stevenson Kenney
September 9, 2002
C S
T
Signal Integrity ADC-LUT-DAC Path
10MHz Sine Signal RS-232
10bit/100MHz
12bit/100MHz
ADC EV Board[AD9214](105MHz)
LUT Board
Agilent 1692A 68ch LA
DAC EV Board[AD9765](125MHz)
PC
I
Q
TDS 3054 DPO(500MHz, 5GS/s)
Agilent 33120AF/AWG (15MHz)
Testbed diagram
LUTin
LUTout
Linear function loaded into LUT
36
J. Stevenson Kenney
September 9, 2002
C S
T
Signal Integrity ADC-LUT-DAC Path (cont.)
(3-a) 10MHz input, No filtering
(3-b) 10MHz input, Filtering (fc=20MHz)
(3-c) Glitch due to the crosstalk of data lines and overflow
37
J. Stevenson Kenney
September 9, 2002
C S
T
Outline
• Introduction to Predistortion• LUT Updates using Sub-sampling
Receivers• RF Envelope Predistortion• Summary and Conclusions
38
J. Stevenson Kenney
September 9, 2002
C S
T
Summary and Conclusions
• Sub-sampling architecture developed to reduce sampling and processing requirements– Sample rate ≥ 2 x BW of input signal– Memory effects may increase this
• RF Envelope pre-D is an alternative to baseband digital pre-D– No digital I/Q stream required– Lower power, high BW