Optimization of PA Linearity and Efficiency through Loadpull Measurements and Simulations using Modulated Signals Onno Kuijken, Komo Sulaksono, Rob Heeres,

Optimization of PA Linearity and Efficiency through Loadpull Measurements and Simulations using Modulated Signals

Onno Kuijken, Komo Sulaksono, Rob Heeres, Léon van den Oever, Luc de MaaijerPhilips Semiconductors, Nijmegen, The NetherlandsBU Mobile & Personal

Semiconductors 2

2 o

f 1

Slid

es

Outline

• Introduction – Problem statement and background• Loadpull measurements using CW and modulated RF

signals• Linearity simulations using modulated RF signals and

behavioural modelling• Conclusions

Semiconductors 3

3 o

f 1

Slid

es

Problem statement

1. Target is to make a linear, efficient power amplifier for WiBro uplink transmissions

2. Implementation platform is laminate-based multi-technology SiP with integrated RF passives

Semiconductors 4

4 o

f 1

Slid

es

Linearity efficiency trade-off

• A WiBro uplink signal is an OFDM(A) signal with up to 864 subcarriers, 2300 MHz < fRF < 2390 MHz.– CCDF on the right– 10 dB PAPR

• Do we really need to back-off by 10 dB?– Very detrimental for

power efficiency

CCDF

0.000001

0.00001

0.0001

0.001

0.01

0.1

1

0 5 10 15

CCDF

Semiconductors 5

5 o

f 1

Slid

es

Multi-technology SiP assembly platformApplication example in (quadband GSM) PA module BGY284

PASSI™ dies for output matching and harmonics filtering, flipchip-mounted

GaAs HBT dies for PA output stages, wirebonded

QUBiC die for PA first and intermediate stages, for biasing and for control, wirebonded

SMDs for low-frequency supply decoupling and for interstage matching circuits, soldered

Laminate substrate as mechanical carrier, I/O rerouting and collector feedlines

Molded package (not shown)

Layout of the two PASSI™ dies and their environment

Semiconductors 6

6 o

f 1

Slid

es

Implementation: PASSI™ Passive Integration on High-Ohmic Silicon

• Pro(during manufacturing)– Reproducible

manufacturing– Stable, reliable,

reproducible assembly

• Con(during product

development)– 3 – 4 weeks processing

leadtime– Little / no post-assembly

tuning

Total 5 mask layers

high ohmic Si

5 m Al

MIM capacitor

425 nm SiNx dielectric

145 pF/mm2

Inductor

Q > 30 @ 2 GHz

L = 1 … 10 nH

Semiconductors 7

7 o

f 1

Slid

es

Outline

• Introduction – Problem statement and background• Loadpull measurements using CW and modulated RF

signals• Linearity simulations using modulated RF signals and

behavioural modelling• Conclusions

Semiconductors 8

8 o

f 1

Slid

es

Loadpull contours with an unmodulated (CW) RF signal

•light blue: Pout contours

•yellow: GT contours

•purple: PAE contours

Impedances are around 2

Z0 = 5

measured at collector

PAX2xV1-1 3.3V 12.5mA*8 LP1t 1/8 2300MHz

0

2

4

6

8

10

12

14

12 16 20 24 28 32 36

Pout [dBm]

Gt

[dB

]

0

10

20

30

40

50

60

70

80

PA

E [

%] Gt_dB

Compr

_1

Eff_%

Gain, compression, efficiency versus output power for different load impedances

• 2.16 – j2.54

• 1.56 – j3.24

• 0.95 – j3.57

Semiconductors 9

9 o

f 1

Slid

es

Z0 = 5

Loadpull contours with a modulated RF signalAt collector of output transistor, for 25 dBm output power

•yellow : ACLR-1 contours Optimum at (3.2 + j2.5) , –37 dBc

•purple : ACLR-2 contours Optimum at (3.2 + j1.6) , –45 dBc

•light blue : PAE contours Optimum at (6.8 + j5.8) , 33.3%

white: ‘point of infinite happiness’, best trade-off overall (4.8 + j2.9) :

ACLR-1 = –34.8 dBc, ACLR-2 = –42.5 dBc, PAE = 27.0%

Semiconductors 10

10

of 1

Slid

es

Conclusion of loadpull measurements

• CW loadpull measurements based on signal statistics and simple amplifier metrics (CCDF, compression point) give values for load impedance which result in non-optimum efficiency

For non-constant-envelope signals, loadpull measurements require real-life stimuli

Semiconductors 11

11

of 1

Slid

es

Outline

• Introduction – Problem statement and background• Loadpull measurements using CW and modulated RF

signals• Linearity simulations using modulated RF signals and

behavioural modelling• Conclusions

Semiconductors 12

12

of 1

Slid

es

reference branch

‘realistic’ behavioural model

piecewise linear model

Simulation set-upFor Ptolemy / ADS co-simulation with modulated RF signals

signal sources (file-based, downloadable to AWG)

Semiconductors 13

13

of 1

Slid

es

Behavioural models-parameters for output match derived from EM simulation

s-parameter dataset block representing PASSI™ output match

derived from Momentum (EM) simulation

Semiconductors 14

14

of 1

Slid

es

Simulation results using behavioural modelsLarge-signal s-parameter representation (measured / simulated)

simulated and measured gain curves

simulated and measured phase curves

ACLR-1 ACLR-2

Simulated –34.4 dBc –45.2 dBc

Measured –33.6 dBc –43.5 dBc

Semiconductors 15

15

of 1

Slid

es

Outline

• Introduction – Problem statement and background• Loadpull measurements using CW and modulated RF

signals• Linearity simulations using modulated RF signals and

behavioural modelling• Conclusions

Semiconductors 16

16

of 1

Slid

es

Conclusions

• Loadpull with modulated excitation signals is a necessary tool in designing highly linear, efficient power amplifiers

• Results obtained have been reproduced by simulations based on behavioural models. The behavioural models can be derived through simulations (including EM), measurements, or a combination

• This leads to almost “first-time right” design for integrated passive circuits