signal processing for power amplifiers - wocc michael luddy.pdf · 4 the need for linearization pa...

1

Signal Processing For Power Amplifiers

Michael Luddy4/23/2005

2

Outline

Motivation and RequirementsCost impact

Architecture and AlgorithmsCFRDPD & MECMEQ

SimulationEnvironmentResults

Emerging Solutions and Future DirectionsDesign DirectionsConclusions

3

Motivation

Motivation and RequirementsCost impact

Architecture and AlgorithmsCFRDPD & MECMEQ

SimulationEnvironmentResults

Emerging Solutions and Future DirectionsDesign DirectionsConclusions

4

The Need For LinearizationPA performance effects Carrier ExpensesDigital Linearization techniques enable:

CapEx Reduction due to lower cost BTS (10-15%)Remove gain stages, circuit hand tuning

OpEx reduction due to higher efficiency PACurrent designs lose ~85% of PA power as heat Annual electrical costs avg ~$2500 per BTS

Performance targetsDouble efficiency of PA (from 12% to 25%) Increase ACLR by > 30 dB

Better than -45 dBc ACLRImprove OBO by more than 4 dBLess than 17.5% EVM

5

Architecture and Algorithms

Motivation and RequirementsCost impactHigher density systems

Architecture and AlgorithmsCrest Factor Reduction CFRDigital Pre-Distortion & Memory Effects Comp DPD & MECModulator Equalization MEQ

SimulationEnvironmentResults

Emerging Solutions and Future DirectionsDesign DirectionsConclusions

6

Signal Processing Features

Crest factor reduction (CFR)Provides more than 6dB improvement in PAR by reducing peak excursionsGoodness by measuring EVM, PCDE and ACLR

Digital pre-distortion (DPD) Correction of gain non-linearityMemory effects (temperature) correction

Modulator Equalization (MEQ)corrects impairments in analog modulator

Phase and gain imbalance, DC offsets

7

Architecture Overview

BPF

Digital Signal Processing

Multicarrier combiner

Crest Factor Reduction

Digital Pre-Distortion

Memory effects compensation

Modulation Equalization

Adaptation Engine

Error Measurement

From

bas

eban

d

Mea

sure

men

t AD

Cs

Sign

al P

ath

DA

Cs

PA

BPF

From LOTh

erm

o A

DC

8

CFR Principles

Comparing WithClipping Threshold

Clipping Filtering

Crest Factor Reduction

Signal from MCC Signal to INF

≤

>

When magnitude exceeds threshold, CFR is performed.The basis of most CFR algorithms is clipping + filtering

][)max((log10 2

2

10 xExCF =

][][

2

2

rEeEEVM RMS =⎟⎟

⎠

⎞⎜⎜⎝

⎛=

][])[max(log10 2'

2

10 rEeEPCDE k

9

Overview of CFR Algorithms

Low complexity.LowLowLowCarrier Phase

Alignment

Peak regrowth is the main problem.LowLowLowError Shaping

The phase distortion should be dynamically distributed among different carriers.

LowLowHighDynamic Phase

Distortion

Similar to clipping and filtering.MediumMediumLowPeak Cancellation

Window length a compromise between ACLR and EVM.

MediumMediumMediumPeak Windowing

Simplest technique.MediumLowHighClipping

COMMENTSPCDEEVMACLRAlgorithm

10

Memory EffectsSlow memory effects

Supply voltage variationAgingAmbient temperatureChannel switching

Fast memory EffectsFast memory effects refer to those which occur so fast that we can not correct them with an adaptation (e.g. LMS) of a predistortion table

11

Digital Predistortion

Vin Vpred

Vin

Vout

Vin Vpred Vout

PAPredistorter

PAVDD

Maximum

improvement

level

12

Digital Predistortion

Look Up Table for slow memory effectsPolynomial for fast memory effectsTraining at startup followed by adaptationDPD LUT uses simple well known techniques

LMS: Well understood, converges for monotonic non-linearity'sRLS: For faster convergence

Adaptation rate (~us) = impairmentsAdaptation engine operates on decimated signal

13

Fast Memory Effects

Fast memory effects create a floor where DPD becomes ineffective, hence we predict thermally induced distortion

Prediction provides correction faster then LUT Prediction error measured in real time and improved in non-real time

Initial correction based on initial calibrationModeled with a polynomial to predict gain compressionDie temperature determined by signal envelope

14

Modulator Equalization

Simplifies analog/IF designCorrects for modulator and DAC imperfections

Gain and phase imbalance DC offset

Gradient descent for 6 parametersInitial calibration the adaptation during system operation

15

Measurement Interface

Measures AM/AM and AM/PM distortionOperates at Fcomposite

Uses generalized sampling theorem [Zhou]Low cost ADC

Hardware decimation and averaging of correction signal

16

Simulation

MotivationCost impactHigher density systems

Architecture and AlgorithmsCFRDPD & MECMEQ

SimulationEnvironmentResults

Emerging Solutions and Future DirectionsDesign DirectionsConclusions

17

Simulation Environment

4 carrier WCDMAAgilent ADS simulation platform

Digital simulation in PtolemyCo-simulated sequential algorithms with MatLab RF amplifier modeled with eesof (AET)

Adaptive peak windowing (Matlab)Bit accurate models using ~14 – 16 bitsSimple adaptation algorithm (MSE)

18

CFR Results

Signal Range (dB)

CCDF

Spectrum

Frequency (GHz)

Original WCDMA signal

After clipping

After clipping + filtering

19

DPD Results

20

Emerging Solutions and Future Directions

MotivationCost impactHigher density systems

Architecture and AlgorithmsCFRDPD & MECMEQ

SimulationEnvironmentResults

Emerging Solutions and Future DirectionsDesign DirectionsConclusions

21

Design Directions

High levels of digital integration (following Moore’s law) are possible thus allowing improvements in system performance with complex, but low cost and low power digital circuitry.Synergistic engineering at the module level enables these promise of higher linearity and efficiencies.

22

Conclusion

Total system design requires skills from packaging, RF design, Materials scientist and DSP designersHigh linearity and efficiency is achievableNew applications can benefit from DPD technology“Old” architecture can have new lives

23

Thank You!