Practical Digital Pre-Distortion Techniques for PA ... · PDF filepractical applications because of its large number of coefficients. ... Signal Training to derive the Memory Polynomial

Practical Digital Pre-Distortion

Techniques for PA Linearization

in 3GPP LTE

Copyright Agilent Technologies 2010SystemVue DPD

Jinbiao XU

May 26, 20101

Jinbiao Xu

Agilent Technologies

Master System Engineer

Agenda

• Digital PreDistortion----Principle

• Crest Factor Reduction

• Digital PreDistortion Simulation

• Digital PreDistortion Hardware Verification


Jinbiao XU

May 26, 20102

Digital Pre-Distortion----- Principle

outP

Linear ResponseOutput Power

SaturationPsat

pd-outP Operating region

with predistortion

Operating region

without predistortion

Input Power

θθθθo-pd = k θθθθout

Input Power

Output Phase

Pi Pi-pd

Desired Output Linear Output


Jinbiao XU

May 26, 20103

Digital Pre-Distortion----- Principle

The DPD-PA cascade attempts to combine two nonlinear systems into one linear result

which allows the PA to operate closer to saturation.

The objective of digital predistorter is to have , where C is a constant.)()( tCxty ≈

4


Jinbiao XU

May 26, 2010

The most important step is to extract PA nonlinear behavior accurately and efficiently.

Memory Polynomial Algorithm

• As the signal (such as 3GPP LTE) bandwidth gets wider, power amplifiers

begin to exhibit memory effects. Memoryless (LUT) pre-distortion can achieve

only very limited linearization performance.

• Volterra series is a general nonlinear model with memory. It is unattractive for

practical applications because of its large number of coefficients.

• Memory polynomial reduces Volterra’s model complexity. It is interpreted as a

special case of a generalized Hammerstein model. Its equation is as follows:

K is Nonlinearity order and Q is Memory order∑∑= =

−−−=

K

k

Q

q

k

kq qnyqnyanz1 0

1)()()(

Copyright Agilent Technologies 2010

5

Memory Polynomial Structure Polynomial Structure

SystemVue DPD

Jinbiao XU

May 26, 2010

= =k q1 0

Signal Training to derive the Memory Polynomial

1. Pre-distorter training:

Nonlinear coefficients are

extracted from the PA

input and PA output

waveforms (ie – on real

physical behavior)

2. Copy of PA : The DPD

model accurately captures model accurately captures

the nonlinearity with

memory effects


6

zUUUa HH 1)(ˆ −=

[ ]TNzzzz )1(,),1(),0( −= K [ ]KQQK uuuuU ,,,,,, 1010 LLL=

[ ]Tkqkqkqkq Nuuuu )1(,),1(),0( −= L

1)()(

)(

−−−

=

k

kqG

qny

G

qnynu

Memory Polynomial Coefficients

[ ]TKQQK aaaaa ˆ,,ˆ,,ˆ,,ˆˆ1010 LLL=

SystemVue DPD

Jinbiao XU

May 26, 2010

• Spectrally efficient wideband RF signals may have PAPR >13dB.

• CFR preconditions the signal to reduce signal peaks without significant

signal distortion

• CFR allows the PA to operate more efficiently – it is not a linearization

technique

• CFR supplements DPD and improves DPD effectiveness

• Without CFR and DPD, a basestation PA must operate at significant

back-off from saturated power to maintain linearity. The back-off

Crest Factor Reduction (CFR) Concepts

back-off from saturated power to maintain linearity. The back-off

reduces efficiency

Benefits of CFR

1. PAs can operate closer to saturation, for improved efficiency (PAE).

2. Output signal still complies with spectral mask and EVM specifications


7

SystemVue DPD

Jinbiao XU

May 26, 2010

Crest Factor Reduction (CFR) Concepts

If you can reduce the Peak-to-Average Ratio of the signal, then for a given

amplitude Peak, you can raise the Average power (up & to the right, above)

with no loss in signal quality.

Thus, CFR enables higher PA efficiency by reducing the back-off, often by 6dB


8

SystemVue DPD

Jinbiao XU

May 26, 2010

Crest Factor Reduction for Multiple-Carrier Signals

• Multiple-Carrier Signals (such as GSM, WCDMA, WiMAX) already have high

PAPR.

• In the future, they will also include multiple waveforms (ie - LTE with 3G WCDMA).

• Therefore CFR will increase in importance for Multi-Carrier PA (MCPA)

linearization.


9

CFR algorithm for multiple carrier signals• PW (Peak Windowing)-CFR

• NS (Noise-Shaping) -CFR

• PI (Pulse Injection)-CFR

• PC (Peak Cancellation)-CFR

SystemVue DPD

Jinbiao XU

May 26, 2010

CFR for 3GPP LTE DL OFDM Signal

• Controls EVM and band limits in the frequency domain.

• Constrains constellation errors, to avoid bit errors.

• Constrains the degradation on individual sub-carriers.

• Allows QPSK sub-carriers to be degraded more than 64 QAM sub-

carriers.

• Does not degrade reference signals, P-SS and S-SS.

• All control channels (PDCCH, PBCH, PCFICH and PHICH) adopts

QPSK threshold.


10

SystemVue DPD

Jinbiao XU

May 26, 2010

LTE CFR (Crest Factor Reduction)

Simulation Results

FFT

FreqSequence=0-pos-neg

Direction=InverseSize=4096 [DFTSize]

FFTSize=4096 [DFTSize]ifft2

0+0*j

Value=0 [0+0* j]zeros

Gain=1

G3

Gain=1

G2

A

BlockSizes=1;300;3495;300 [[1,Half_UsedCarriers,DFT_zeros,Half_UsedCarriers]]

A3

A

BlockSizes=300;300 [[Half_UsedCarriers, Half_UsedCarriers]]

A2

FFT


Direction=ForwardSize=4096 [DFTSize]

FFTSize=4096 [DFTSize]

fft

FFT


Direction=InverseSize=4096 [DFTSize]

FFTSize=4096 [DFTSize]

ifft1

0+0*j

Value=0 [0+0* j]

DC

DPD_Rad ius Clip

input out put

ClippingThreshold=16.5e-6 [ClippingThreshold]

DPD_RadiusClip

DPD_LTE_CFR_Post Proc

input

ref

SC_St at us

Qm

out put

OutOfBandAlgorithm=Armstrong algorithm

EVM_Threshold_64QAM=0.1 [EVM_Threshold_64QAM]EVM_Threshold_16QAM=0.1 [EVM_Threshold_16QAM]

EVM_Threshold_QPSK=0.1 [EVM_Threshold_QPSK]

SSS_Ra=0 [SSS_Ra]PSS_Ra=0 [PSS_Ra]

UEs_Pa=0;0;0;0;0;0 [UEs_Pa]

PDSCH_PowerRatio=p_B/p_A = 1 [PDSCH_PowerRatio]PDCCH_Rb=0 [PDCCH_Rb]

PDCCH_Ra=0 [PDCCH_Ra]PBCH_Rb=0 [PBCH_Rb]

PBCH_Ra=0 [PBCH_Ra]

PHICH_Rb=0 [PHICH_Rb]PHICH_Ra=0 [PHICH_Ra]

PCFICH_Rb=0 [PCFICH_Rb]

RS_EPRE=-25 [RS_EPRE]OtherUEs_MappingType=0;0;0;0;0 [OtherUEs_MappingType]

UE1_MappingType=0;0;0;0;0;0;0;0;0;0 [UE1_MappingType]CyclicPrefix=Normal [CyclicPrefix]

OversamplingOption=Ratio 4 [OversamplingOption]

Bandwidth=BW 10 MHz [Bandwidth]D1

LTE Downlink 10MHz,

Sampling Rate 61.44MHz,

QPSK,

EVM threshold 10%

Simulation Results


Jinbiao XU

May 26, 201011

DPD Simulation Workspace

Step 1 is to Generate Waveform for DPD

Step 3 is for DUT Model Extraction

Step 4 is for DPD Response

Compared with hardware verification tool,

simulation tool does not include Step 2

and Step 5.

Hardware verification toll will be

introduced later.


Jinbiao XU

May 26, 201012

LTE DPD simulation for a memoryless nonlinear PA

MathPow10

FunctionType=Pow10

M12 {Math@Data Flow Models}

G28 {Gain@Data Flow Models} G23 {Gain@Data Flow Models}

DPD_PAModel

PA_IN PA_OUT

DPD_PAModel_1

T

SampleRate=122.9e+6Hz [SamplingRate]S4

Fc

EnvCx

Fc=2GHzC3

DPD_PreDistorterDPD_Input

DPD_Coef

DPD_Output

NumOfInputSamples=61440 [NumOfInputSamples]NonlinearOrder=9MemoryOrder=7

D1

Spectrum Analyzer

SegmentTime=50µs

Start=0sMode=TimeGate

AfterDPD

Fc

CxEnv

E1

Re

Im

R6

Periodic=YESFile='Step3_DPD_Coefficients_Imag.txt

R5

Periodic=YES

File='Step3_DPD_Coefficients_Real.txtR4

T

SampleRate=122.9e+6Hz [SamplingRate]

S1

Fc

EnvCx

Fc=2e+9Hz [FCarrier]

C2

Spectrum Analyzer

SegmentTime=50µsStart=0s

Mode=TimeGateAfterDPD_PA

Gain=0 [mPowers(1)]

G18 {Gain@Data Flow Models}

MathPow10

FunctionType=Pow10


MathPow10

FunctionType=Pow10


MathPow10

FunctionType=Pow10


MathPow10

FunctionType=Pow10


MathPow10

FunctionType=Pow10M8 {Math@Data Flow Models}

MathPow10

FunctionType=Pow10


MathPow10

FunctionType=Pow10


MathPow10


Gain=2 [mPowers(3)]


Gain=1 [mPowers(2)]


Gain=4 [mPowers(4)]G15 {Gain@Data Flow Models}

MathPow10


Gain=31.623 [mPars(1)]


Gain=-72.068 [mPars(2)]


Gain=254.726 [mPars(3)]


Gain=-1107.963 [mPars(4)]G6 {Gain@Data Flow Models}

Gain=2956.358 [mPars(5)]


Gain=-4462.485 [mPars(6)]


Gain=3782.968 [mPars(7)]G9 {Gain@Data Flow Models}

Gain=-1653.647 [mPars(8)]


Gain=281.85 [mPars(9)]


A1 {Add@Data Flow Models}

MathPow10


FunctionType=Pow10

MathPow10


MathPow10

FunctionType=Pow10


MathPow10

FunctionType=Pow10


Gain=6 [pPowers(5)]


Gain=4 [pPowers(4)]


Gain=2 [pPowers(3)]


Gain=1 [pPowers(2)]


Gain=0 [pPowers(1)]G28 {Gain@Data Flow Models}

Gain=-1.159 [pPars(1)]G23 {Gain@Data Flow Models}

Gain=0.917 [pPars(2)]


Gain=-1.746 [pPars(3)]


Gain=0.992 [pPars(4)]


Gain=-0.155 [pPars(5)]

G19 {Gain@Data Flow Models} A2 {Add@Data Flow Models}

Gain=14 [mPowers(9)]


Gain=12 [mPowers(8)]


Gain=10 [mPowers(7)]G12 {Gain@Data Flow Models}

Gain=8 [mPowers(6)]


Gain=6 [mPowers(5)]


Mag

Phase

C2 {CxToPolar@Data Flow Models}

MathLog10

FunctionType=Log10M17 {Math@Data Flow Models}

MathLog10

FunctionType=Log10M2 {Math@Data Flow Models}

LimiterType=linear

Top=0.788 [mXmaxVolt]

Bottom=0K=1

L2 {Limit@Data Flow Models}

LimiterType=linear

Top=0.86 [pXmaxVolt]

Bottom=0K=1

L3 {Limit@Data Flow Models}

M1 {Mpy@Data Flow Models}

Gain=1G29 {Gain@Data Flow Models}

Mag

Phase

P1 {PolarToCx@Data Flow Models}

Bus=NO

Data Type=Complex

PA_OUT {DATAPORT}

A3 {Add@Data Flow Models}

10e-201

Value=1e-200

C1 {Const@Data Flow Models}

Bus=NOData Type=Complex

PA_IN {DATAPORT}

EVM (dB)

ACLR (dB)


Jinbiao XU

May 26, 201013

LTE DPD simulation for a nonlinear PA with memory

DPD_PreDis to rterDPD_I nput

DPD_Coef

DPD_O ut put

NumOfInputSamples=61440 [NumOfInputSamples]

NonlinearOrder=9

MemoryOrder=7

D1

Fc

CxEnv

E4

Amplifier

MaxFallTC=16.28e-9s [MaxFallTC]

MaxRiseTC=16.28e-9s [MaxRiseTC]

PdBmMaxMem=10 [PdBmMaxMem]

PdBmNoMem=-30 [PdBmNoMem]

dBc1out=10dBm

GCType=none

NoiseFigure=0

Gain=30 [GaindB]

GainUnit=dB

Amplifier1

T

SampleRate=122.9e+6Hz [SamplingRate]

S1

Fc

EnvCx

Fc=2e+9Hz [FCarrier]

C2

Spectrum Analyzer

SegmentTime=50µs

Start=0s

Mode=TimeGate

AfterDPD_PA

Fc

EnvCx

Fc=0.2e6Hz

C8

Re

Im

R6

Periodic=YES

File='Step3_DPD_Coeffic ients_Imag.tx t

R5

Periodic=YES

File='Step3_DPD_Coeffic ients_Real.tx t

R4

Fc

CxEnv

E1

Fc

CxEnv

Mag

Phase

Mag

Phase

Fc

EnvCx

Bus=NO

Data Type=Floating Point (Real)

control {DATAPORT}

output {DATAPORT}

Bus=NO

Data Type=Envelope Signal

input {DATAPORT}

RiseFallTCv in v out

RefR=50O [RefR]MaxFallTC=100e-6s [MaxFallTC]

MaxRiseTC=10e-6s [MaxRiseTC]PdBmMaxMem=30 [PdBmMaxMem]

PdBmNoMem=10 [PdBmNoMem]

R1 {RiseFallTC@SV_VC_TC Models}

Amplifier Amplifier

EVM (dB)

ACLR (dB)

Fc

E1 {EnvToCx@Data Flow Models}

Phase

C3 {CxToPolar@Data Flow Models} P1 {PolarToCx@Data Flow Models}

Fc

Fc=0.2e6Hz

C1 {CxToEnv@Data Flow Models}

Bus=NO

Data Type=Envelope Signal

output {DATAPORT}

NoiseFigure=0 [NoiseFigure]

Gain=1A1 {Amplifier@Data Flow Models}

Gain=1 [Gain]

GainUnit=voltage [GainUnit]A2 {Amplifier@Data Flow Models}


Jinbiao XU

May 26, 201014

DPD Hardware Verification Flowchart

Create DPD Stimulus

Capture DUT Response

DUT Model Extraction

DPD Response

DPD HW Flowchart consists of 5 steps:

• Step 1 (Create DPD Stimulus) is to

download waveform (LTE or User defined)

into ESG/MXG.

• Step 2 (Capture DUT Response) is to

capture both waveforms before power

amplifier and after power amplifier from

PSA/MXA/PXA by using VSA89600

software.

• Step 3 (DUT Model Extraction) is to extract

PA nonlinear coefficients based on both

Verify DPD Response

captured PA input and PA output waveforms

and then to verify DPD by using PA

nonlinear coefficients.

• Step 4 (DPD Response) is to download the

waveform (LTE or User Defined) after pre-

distorter (by using PA nonlinear coefficient

from Step 3) into ESG/MXG, this real signal

passes through the PA DUT, capture PA

output waveform from PSA/MXA/PXA by

using VSA89600 software.

• Step 5 (Verify DPD Response) is to show

the performance improvement after DPD.


Jinbiao XU

May 26, 201015

DPD Hardware Verification Workspace Structure


Jinbiao XU

May 26, 201016

DPD Hardware Verification Platform

10MHz Reference

External Trigger

1. PA input signal capture

Signal source:

LTE 10MHzAgilent MXG/ESG PSA/MXA/PXA

10MHz Reference

External Trigger

Attenuator

2. PA output signal capturePSA/MXA/PXAMXG/ESG


Jinbiao XU

May 26, 201017

DPD Hardware Verification – LTE (Step 1)

Step 1: Create Stimulus The CFR must be enable in LTE

source.

LTE paramters (such as bandwidth,

Resource Block allocation and etc)

can be set.

The download waveform transmit

power, length also can be set.


Jinbiao XU

May 26, 201018

Step 2: Capture DUT Response


Firstly, connect the ESG directly

with the PSA/PXA and click the

“Capture Waveform” button in the

“Capture PA Input” panel in the

GUI. The captured signal is the

input of the PA DUT.

Then, connect the ESG with the

DUT, and then connect the DUT DUT, and then connect the DUT

with the PSA/PXA and click the

“Capture Waveform” button in the

“Capture PA Output” panel in the

GUI. The captured signal is the

output of the PA DUT.

These I/Q files are stored for

further usage.


Jinbiao XU

May 26, 201019

Step 3: DUT Model Extraction


This step is to extract

PA nonlinear

coefficient from the

PA input and PA

output waveform and

get the coefficients of

the DPD model.

DPD Verification AM-AM


Jinbiao XU

May 26, 201020


Step 4: DUT Response This step is to apply the DPD model extracted in

Step 3. The generated LTE downlink signal is

firstly pre-distorted by the extracted model, and

then downloaded into the ESG.


Jinbiao XU

May 26, 201021


Spectrum

EVM

ACLR

Step 5: Verify DUT Response

This step is to verify the performances

of the DPD (including spectrums of the

DUT output signal w/ and w/o DPD,

EVM and ACLR).

EVM (dB)

ACLR (dB)


Jinbiao XU

May 26, 201022

Hardware Verification Results of Doherty PA

EVM (dB)

ACLR (dB)


Jinbiao XU

May 26, 201023

1. Lei Ding, Zhou G.T., Morgan D.R., Zhengxiang Ma, Kenney J.S., Jaehyeong Kim, Giardina C.R., “A robust digital

baseband predistorter constructed using memory polynomials”, Communications, IEEE Transactions on, Jan. 2004,

Volume: 52, Issue:1, page 159-165.

2. Lei Ding, “Digital Predistortion of Power Amplifiers for Wireless Applications”, PhD Thesis, March 2004.

3. Roland Sperlich, “Adaptive Power Amplifier Linearization by Digital Pre-Distortion with Narrowband Feedback using

Genetic Algorithms”, PhD Thesis, 2005.

4. Helaoui, M. Boumaiza, S. Ghazel, A. Ghannouchi, F.M., “Power and efficiency enhancement of 3G multicarrier

amplifiers using digital signal processing with experimental validation”, Microwave Theory and Techniques, IEEE

Transactions on, June 2006, Volume: 54, Issue: 4, Part 1, page 1396-1404.

5. H. A.Suraweera, K. R. Panta, M. Feramez and J. Armstrong, “OFDM peak-to-average power reduction scheme with

spectral masking,” Proc. Symp. on Communication Systems, Networks and Digital Signal Processing, pp.164-167, July

2004.

References

2004.

6. Zhao, Chunming; Baxley, Robert J.; Zhou, G. Tong; Boppana, Deepak; Kenney, J. Stevenson, “Constrained Clipping for

Crest Factor Reduction in Multiple-user OFDM”, Radio and Wireless Symposium, 2007 IEEE Volume , Issue , 9-11 Jan.

2007 Page(s):341- 344.

7. Olli Vaananen, “Digital Modulators with Crest Factor Reduction Techniques”, PhD Thesis, 2006

8. Boumaiza, et a, “On the RF/DSP Design for Efficiency of OFDM Transmitters” , IEEE Transactions on

Microwave Theory and Techniques, Vol. 53, No. 7, July 2005, pp 2355-2361.

9. Boumaiza, Slim, “Advanced Memory Polynomial Linearization Techniques,” IMS2009 Workshop WMC

(Boston, MA), June 2009.

10. Amplifier Pre-Distortion Linearization and Modeling Using X-Parameters, Agilent EEsof EDA


24

SystemVue DPD

Jinbiao XU

May 26, 2010

Documents

Practical Digital Pre-Distortion Techniques for PA ... · PDF filepractical applications because of its large number of coefficients. ... Signal Training to derive the Memory Polynomial