Nortel Networks Institute University of Waterloo · University of Waterloo, Waterloo, ON, Canada ... SDR poses several challenging design factors that call for Key Enabling Technologies

Nortel Networks InstituteUniversity of Waterloo

Dr. Slim Boumaiza

EmRG Research GroupElectrical and Computer Engineering,

University of Waterloo, Waterloo, ON, Canada

EmRG Research Group, Electrical and Computer Engineering www.ece.uwaterloo.ca\~sboumaiz

Advanced Techniques in Power Efficiency

and Linearity Enhancement of 3G and Beyond Wireless Transmitters

http://www.ece.uwaterloo.ca/~sboumaiz



• Wireless Communication Evolution & Challenges

• SDR Key Enabling Technologies

– Ultra Linear Front-End

• Linearization techniques

• Digital Predistortion

– High Efficiency Front End

• Doherty Amplifier

• Linear Amplification Using Nonlinear Components

• Conclusion

Outline





Overview on Wireless Communication Evolution & Challenges

Wireless Communication Evolution

CDMA

GSM Edge

3GPP

TD-SCDMA

WiMAXLTE

PAR

LowFairHigh

New standards boost Spectral Efficiency bits/s/Hz improves thanks to more advanced modulation schemes: up to 100MB/s capability targeted by LTE

RF signal Peak to Average Ratios typically: 3.4dB Edge / 6.5dB W-CDMA / 10dB WiMAX

Infrastructure Deployment Phase OutStandardization






-100

-90

-80

-70

-60

-50

-40

-30

-20

2130 2135 2140 2145 2150

PA output

Original signal

PS

D (

dB

m)

Frequency (MHz)

The out-of-band noise increases and affects the adjacent channels

(ACPR)

The in-band noise increases and affects the transmission quality

(EVM)

Wireless Communication Challenges




Dollars in CapEx and OpEx:• utility costs (power for PA and

cooling)

• space (cost of rent)

• racks, heat sinks, fans, etc.

• number of PAs required


Very Stringent Linearity Requirements for RF front end

Operation in Large Power Backoff Region good linearity but poor Efficiency

Linearization to enhance the linearity vs. efficiency tradeoff is more critical than ever

EfficiencyTranslates

Better Reliability is inversely

proportional to heat-load





Standards operate at diverse frequencies with a more and more higher median frequency as

available spectrum is scarce

GSM and CDMA already deployed

WiMAX in “design” stages

3G & 3G Evolution (HSDPA deployment, LTE on the

horizon, China (TD-SCDMA))


Median frequency is evolving: 800/900MHz 5 years ago, 2.1 GHz now , 3+ GHz in the future





These different standards will go on coexisting

Broad spectrum of base stations and terminals has to be maintained to cover all operator’s needs

Market evolution calls for flexible infrastructure to enable dynamic networks management and infrastructure sharing

Solution: Software Defined Radio where one flexible architecture for both base stations and terminals is used to meet the different requirements






An

ten

na

s

Multi-band RFTransmitter

Multi-band RF Receiver

Basebandanalogue

& digital proc.

Higher LayerFunctions

control

Software Defined Radios Generic Diagram






SDR poses several challenging design factors that call for Key Enabling Technologies

New device technology

New Design Methodologies: RF and

Digital Mixed Design

Multi-band RF Sub-Blocks

(Antenna, Power

Amplifiers,Filters,

Combiners)

Ultra Linear RF

Front Ends: Advanced Linearizati

on Techniques

High Efficiency RF Front

Ends

Advanced Digital Signal

Processing Algorithms






SDR Key Enabling Technologies: Ultra Linear RF Front-End

Linearization TechniquesOverview





Feedback

Requires high gain and stable analog feedback at the carrier frequency

Close the feedback at baseband Cartesian and Polar

The non-linear inter-modulation products are reduced by the same ratio as the feedback gain.


Feedback has a limited bandwidth capability.

Potential instability due to the propagation delay in the feedback loop.

Its design become challenging at high frequencies.

It is more suitable for an integrated handset transmitter design as the important transmitter delay makes it very difficult for base station




Feedforward

The output of the Power Amplifier is coupled and compared to original input. Finally added to the output of main amplifier.

The delay of two amplifiers have to be replicated precisely

This is an open-loop system which is vulnerable to environmental changes





Feedforward





Predistortion

Basically estimate the nonlinear behavior of PA at the output and tries to modulate the input signal with inverse function.

The complexity of the system might incur significant power overhead





PredistortionThe Predistorter generates non-linearities that cancel amplifier distortions when combined.

It can be performed at RF or at base band using FPGA processor.

Analog RF Predistortion is widely used for slow-slope non-linear amplifiers such as TWTA.

Analog RF Predistortion is a mature technology for mass production.

Good linearity improvement in TWTAs and less significant improvement for MESFET class A and AB.

Does not degrade much the power efficiency of the PA.

Needs an adaptation loop to track the drift of the PA.





Predistortion

DPD is becoming technically mature and an increasing number of RF transmitters and PAs designers are bet on DPD for tackling linearization problems.

Number of published patents and scientific articles

per year on the subject "digital predistortion"

D. Rönnow, Measurement, analysis, modelling and digital predistortion of RF/Microwave power amplifiers, Racomna

Research AB,





Comparaison Predistortion vs.Feedforward vs. Feedback

SSPAs for Base

StationsPAs for Mobile

Stations

TWTAs & SSPAs for

Satcom applications

Intrinsically

non adaptiveIntrinsically

adaptive

Intrinsically non

adaptiveAdaptation

LowHighHigh/HighPower

Efficiency

HighMediumMedium/MediumComplexity

UltraGoodMedium/UltraLinearity

WideNarrowUltra/ MediumFrequency

bandwidth

FeedforwardFeedbackRF Predistortion/

Digital Predistortion





High>100 MHz25 - 35 dBFeedforward

Relative

Cost

Correction

Bandwidth

Correction

Capability*

Linearization

Technique

Medium<5MHz10 - 20 dBEnvelope

Feedback

Low>25MHz5 - 10 dBAnalog Pre-

Distortion

Medium>50MHz10 - 20 dBAdaptive Pre-

Distortion

High>100 MHz25 - 35 dBFeedforward

Relative

Cost

Correction

Bandwidth

Correction

Capability*

Linearization

Technique

Medium<5MHz10 - 20 dBEnvelope

Feedback

Low>25MHz5 - 10 dBAnalog Pre-

Distortion

Medium>50MHz10 - 20 dBAdaptive Pre-

Distortion


Comparaison Predistortion vs.Feedforward vs. Feedback




Digital Predistortion Technique

( )inx t

Pin

G

Pin

Gn

Gn

Pin Pin

G G

Pin

Nonlinear PA/TRx Linearized PA/TRx

Pin

G

Psatin

Psatin

Predistortion Function

( )inx t ( )outy t( )dx t ( )outy t

Predistortion function adjusts the input signal envelope so that it brings in distortions out of phase with those generated by the PA/TRx nonlinearities.

Digital Predistortion Concept





The response of nonlinear circuits can be assessed using different types of excitation signals

• Continuous Wave signal (trace gain in power sweep mode)

• Two tones signal (Inter-modulation products)

• Multi-sine signal (in-phase or randomly distributed phases)

• Modulated signals synthesized according to the targeted standard

Nonlinear Circuit Characterization





46

47

48

49

50

51

52

-25 -20 -15 -10 -5 0 5

CW

WCDMA

8-Tones

Gain

(d

B)

Pin (dBm)

-80

-75

-70

-65

-60

-55

-25 -20 -15 -10 -5 0 5

CW

WCDMA

8-TonesPh

ase

(d

egree

)Pin (dBm)

Measured transmitter’s AM/AM and AM/PM characteristics for various excitation signals

Nonlinear Circuit Characterization





The power amplifier nonlinearity characterization is assessed under realistic signal and not continuous wave

Time-Domain Nonlinear Circuits Characterization Approach





Gain vs. input power measurements with VNA under CW signal, peak power analyzer and the Time domain test bed with CDMA 2000 signal.

Phase compression vs input power measurements with VNA with CW signal

and the Time domain test bed with CDMA 2000 signal.

56.5

57

57.5

58

58.5

-24 -22 -20 -18 -16 -14 -12 -10 -8

VNA mesurements

Peak power analyser

Proposed test bed

Power input (dBm)






To achieve accurate characterization the device under test should be excited as close as possible to the real operation condition

This approach demonstrated and solved the inaccuracy of VNA based Characterization of PA nonlinearity






Amplifier : cascade of MRF21045 and MRF21085 (Freescale semiconductor)

Frequency : 2140 MHz

Output power : 90W Peak

Signal : WCDMA 1-carrier

25 dB ACPR decrease from -35dBc to -60dBc

-100

-90

-80

-70

-60

-50

-40

-30

-20

2130 2135 2140 2145 2150

memoryless DPD

without DPD

Original signal

PS

D (

dB

m)

Frequency (MHz)

The good accuracy of the characterization is demonstrated through the Digital Predistorter performance






-100

-90

-80

-70

-60

-50

-40

-30

-20

2120 2125 2130 2135 2140 2145 2150 2155 2160

sans mémoiresans prédistorsionSignal original

PS

D (

dB

)

Frequence (MHz)Frequency (MHz)

-100

-90

-80

-70

-60

-50

-40

-30

-20

2120 2130 2140 2150 2160

sans mémoiresans prédistorsionSignal original

PS

D (

dB

)

Frequence (MHz)Frequency (MHz)

New phenomenon takes place and neglected before

As the signal bandwidth signal increases the predistortion

capability is reduced






The residual nonlinearity shown by the linearized transmitter can

be explained•Nonlinearity characterization inaccuracy

•Memory effects which is defined as

Frequency domain: *IMD3 products magnitude and phase

changes as a function of the frequency spacing.

Time domain: PA output depends not only on the current input but also on its previous values

)(),(),...,()( Mtxtxtxfty

f1 f2 f1 f22f1-f2 2f2-f1 *IMD3=Amp(2f2-f1)

Memory Effects Investigation






Hypothesis: PA nonlinearity depends only the current input signal amplitude

AM/AM Curve AM/PM Curve





Tx

C

Vgs

C

Vds

Power amplifier schematic

The time domain characterization approach is used once again

to investigate the memory effects

RF Transmitters Block Diagram

PAD/A RF_out

Up

converterDigital

Mod

I

Q






Asymmetry between the upper and lower IMD3 components

Dependency of the IMD3 magnitude and phase on tone spacing

-65

-60

-55

-50

-45

0 5 10 15 20 25

IMD3_L

IMD3_R

IMD

3 (

dB

m)

Frequency spacing (MHz)

What to measure: IMD3 distortion at the transmitters output vs.

tone spacing

IMD3_L=Amplitude(2f1-f2)IMD3_R=Amplitude(2f2-f1)

Two regions are considered:

- Low freq spacing long term memory effects

- Medium to high freq spacing short term memory effects






thR

thC

)(tPdissip

tTjcT

thR

thC

)(tPdissip

tTjcT thC

)(tPdissip

tTjcT

Long-term (electro-thermal) memory effects

attributed to the dynamic changes of the junction

temperature as a function of the input signal strength

variation of transistor electrical parameters

Heat Flux

Equivalent thermal circuit

ChipPackage

Heat sink Tamb

Tj

Tc

1

tt

j th dissip cT t e e R P t T dt K






Long-term memory effects

• IMD3 distortion attributed to thermal effects dependent on

the frequency spacing.

• IMD3 attributed to the PA electrical non-linearity remains

constant when varying the frequency spacing.

tx ty

NL Electrical

PA Behaviour

NL Thermal

Effects

lev ,

lthv ,

rev ,

rthv ,

lv

rv

rerthr

lelthl

vvv

vvv

,,

,,

rerthr

lelthl

vvv

vvv

,,

,,






Short-term (electrical) memory effects

-Attributed to terminals impedance variation of the biasing and matching circuits

as function of frequency

Im(IM3L)

Re(IM3L)

2sd ordre(envelop)

2sd ordre(harmonic)

w1-

w2

w2-

w1

0

2w1-

w2

2w2-

w1

w1

2w1

2w2

w1+

w2

w2 freq

A

Spectrum composition under two tones excitation

Tr POUTPINMatching

circuit

at the carrier frequency and its harmonics

Biasing circuit

at the envelop frequency






• Narrow-band signals

Electro-thermal memory effects are attributed to the dynamic

changes of the junction temperature as a function of the input

signal strength which leads to the transistor electrical parameters

variation (e.g. gain magnitude and phase)

• Wideband signals

Electrical memory effects are due to the variation of terminal

impedances of the biasing and matching circuits impedances over

the input signal bandwidth around- the carrier frequency

- its harmonics

- base band frequencies







• Several approaches were used to develop a DPD that can compensate for the static nonlinearity of the PAs as well as the dynamic distortion due to the memory effects:

– Volterra series: for mild nonlinear case

– Hammerstein and Wiener with less success

– Neural Networks

– Augmented Wiener and Hammerstein

– Memory polynomial which constitute a good approximation of the Volterra series





Memoryless static nonlinear subsystem

( )u n

FIR2

Memory effect subsystem

FIR1

( ) ( )x n x n

( )u n ( )x nAM/AM and

AM/PM LUT

( )y nx +

( ) ( ) ( ) ( ) ( ) ( )i qx n G u n jG u n u n G u n u n

1 21 1

0 0

( ) ( ) ( ) ( )M M

i i

i i

y n a x n i b x n i x n i

Augmented Hammerstein Model


Augmented Weiner/Hammerstein Nonlinear Behavioral Models




-100.0

-90.0

-80.0

-70.0

-60.0

-50.0

-40.0

1.92 1.93 1.94 1.95 1.96 1.97 1.98 1.99 2

Frequency (GHz)

PS

D (

dB

/Hz)

(a)

(b)

(c)

(d)

(a) Without predistorter. (b) With memoryless predistorter. (c) Hammerstein predistorter with a 128-tap FIR filter. (d) Augmented Hammerstein predistorter with two 20-tap filters.

Augmented Weiner/Hammerstein Nonlinear Behavioral Models





Memory Polynomial DPD

Typical Memory Polynomial (MP):

Each branch is a Polynomial Poly:

The General Model Equation is:

Where i is polynomial branch,

P is the polynomial order.

Poly 1

Z-1

Z-1

Z-1

+X y

Poly 2

Poly K)()()(

1

0 1

, inxinxhny

pK

i

P

p

ip

12

321 ...

P

P xxhxxhxxhxhPoly





MEASUREMENT RESULTS

Memory Polynomial


• DPD 2 achieves similar results to DPD 1.

• Number of coefficients required is reduced from 7x10=70 to 7+9x4=43.

Polynomial Order

DPD# 1 2 3 4

Po

lyn

om

ial B

ran

ch

P1 7 7 7 7

P2 7 4 4 2

P3 7 4 4 2

P4 7 4 4 2

P5 7 4 4 2

P6 7 4 2 2

P7 7 4 2 2

P8 7 4 2 2

P9 7 4 2 2

P10 7 4 2 2

4-Carrier WCDMA

PAPR = 7.4dB

Output Spectrum of Linearized 400W PEP Power Amplifier




SDR Key Enabling Technologies:

High Efficiency RF Front-End





Doherty Amplifier

Conventional power amplifiers used in the previous experiments

have limited efficiency since the load impedance is kept constant

as the input power changes.

This load was chosen so that the efficiency is maximized for the

max input power level

Hence, for lower input power the efficiency will drop.

It is advantageous if the output load impedance can be

modulated as function of the input power so that the efficiency is

increased for low power levels

Tr POUTPINMatching

circuitZl

Doherty Amplifier





Doherty Amplifier • Doherty amplifier is based on the inherent load modulation

vs. input power level• For low powers peak PA is off

The carrier PA see a 2Ro load

• For higher powers (6 dB before

saturation) the peak turns on

The actual load decreases up to Ro

Comprehensive design approach suitable for designing a Doherty amplifier is needed.

• The design of the carrier and peak amplifiers relies on an accurate – determination of the optimum loads vs. input power level– large signal transistor model. This can be built based on load

pull measurement data• The model provides the RF output voltage (magnitude and

phase components) and the drain current as a function of input power and load reflection coefficient.


Doherty Amplifier




Load-Pull Test Bench


Doherty Amplifier




The appropriate matching circuits at the output of the carrier and peak amplifiers can be designed to produce the required load impedance displacement at the output of the two transistors

Load impedance displacement of the main amplifier

14W Doherty amplifier prototype


Doherty Amplifier




At 10dB Back off Doherty amplifier allows 10% higher PAE than

the class AB. Max PAE=57%


Doherty Amplifier




Doherty Amplifier Linearization

DPD linearization of 2C-WCDMA signal

ACPR enhancement = 27 dB (final ACPR = -47dBc)

-100

-90

-80

-70

-60

-50

-40

1.92 1.94 1.96 1.98

Frequency (GHz)

PS

D (

dB

m)

W/o Pred.

With Pred.





Linear Amplification with Nonlinear Components (LINC)

LINC Scheme

The idea is to transform the signal into two constant envelop signals and make use of high efficient nonlinear amplifiers

Outphasing concept: Decompose the amplitude and phase varying signal into two constant envelope signals before amplification





• Use of nonlinear power amplifiers (PAs) with very high efficiency

• The PAs are operated in the saturation region but the signal quality is conserved

Ensure linear amplification with high efficiency PAs

• The output signals of the two PAs are combined to recover the amplitude and phase modulated signal

• LINC efficiency is given by the product of three efficiency types:

• The modulation or the combining efficiency is defined by:

1 2

cosoutM

P

P P

LINC PA C M

LINC Scheme





Wilkinson

Combiner

Chireix

Combiner

R

LINC Scheme





LINC Scheme

Wilkinson

(Lossy)

Chireix

(Lossless)

Chireix + jβ

(Lossless)

Isolation

Linearity

efficiency

At variable

back-off


Combining Techniques




LINC Efficiency Degradation

Amplification of high signals that will be combined destructively

High DC power consumption for to generate power samples

Signal peaks’ probability is generally low

significant degradation of power efficiency.

0

20

40

60

80

0 5 10 15 20 25 30

LIN

C e

ffic

ien

cy (

%)

output power (dBm)

LINC efficiency < 7%

for signals with PAPR = 10 dB





0

10

20

30

40

50

60

-20 -15 -10 -5 0 5

LINC using Wilkinson

Class B

Eff

icie

ncy (

%)

Pin

(dBm)

combining efficiency versus power back-off (measurement)

LINC Efficiency Degradation





MM-LINC concept

• MM-LINC stands for mode-multiplexing LINC

• Since the instantaneous efficiency decreases rapidly as input power decreases, this approach proposes to:

– Use the 2 parallel PAs based architecture in balanced mode for high power levels to keep efficiency high

– Use the 2 parallel PAs based architecture in LINC mode for low power levels to keep gain constant in that region

Better linearity than class B or class C PAs

Better efficiency than LINC architecture





MM-LINC Signal Decomposition

0

1 2

0

1

2

11

2

/

in

in

s t r r

s t

s t j e t r r

2

0

2

0

0

1

max

is the threshold to switch from one mode to another

is the normalized threshold value

re t

r t

r

r

r





MM-LINC efficiency performance

Efficiency performance of the MM-LINC technique

compared to LINC and class B (measurement)

0

10

20

30

40

50

60

-20 -15 -10 -5 0 5

LINC

Part-time LINC

Class B

Pow

er a

dd

ed e

ffic

ien

cy

(%

)

Pin (dBm)

MM LINC





MM-LINC Linearity performance

Linearity versus efficiency performance of the MM-LINC

technique compared to LINC and class B (measurement)

0

5

10

15

20

25

30

30

31

32

33

34

35

36

0 0.2 0.4 0.6 0.8 1

EVM

PAE

Pout

EV

M (

%),

PA

E (

%)

Pou

t (dB

m)

LINCClass B

0

5

10

15

20

25

30

30

31

32

33

34

35

36

0 0.2 0.4 0.6 0.8 1

EVM

PAE

Pout

EV

M (

%),

PA

E (

%)

Pou

t (dB

m)

LINCClass B





MM-LINC performance

• The MM-LINC is suitable for mobile applications (low complexity)

• It improves a trade-off between efficiency and linearity depending on the linearity requirements

• MM-LINC has better efficiency and better dynamic range than the traditional LINC

• Limitations :The distortions introduced by the PAs saturation is not

corrected for in this architecture

The linearity of MM-LINC is lower than the linearity of LINC technique when using Wilkinson combiner





• Software Designed Radio Base Station is foreseen to be vital for future wireless networks.

• It causes many challenges and calls for many enabling technologies

• Digital Predistortion Technique using Time Domain Characterization method is needed to achieve Ultra Linear Front Ends

• As signals’ bandwidth is more and more wider, memory effects is a critical design factor.

Conclusion





• The mitigation of their effects can be addressed using DPD with memory.

• Augmented Hammerstein/Weiner and Polynomial behavior models showed good capability in predicting and linearizing the dynamic nonlinear behavior of wideband wireless transmitter

• However, it is advantageous to minimize them during the PA design


Conclusion




• Doherty Amplifier showed substantial power efficiency enhancement

• Further improvement could be achieved using a Digital Asymmetrical Doherty

• Mode Multiplexing LINC achieved very interesting linearity vs. efficiency tradeoff. It could be a good candidate for handset Pas.

• RF/DSP mixed co-design approach will be required to tackle the SDR challenges and meet their stringent requirements


Conclusion




Nortel Networks Institute University of Waterloo · University of Waterloo, Waterloo, ON, Canada ... SDR poses several challenging design factors that call for Key Enabling Technologies

Documents