Progress on LLRF applications development

Zheqiao Geng, Valeri Ayvazyan, Stefan SimrockFLASH Seminar

07.04.2009

2

Outline

Goals of LLRF applicationsApplications studied in January 2009

DAC DC offset calibration for LLRF controllerCavity quench detectionAdaptive feed forward

Summary

3

Goals of LLRF applicationsLLRF applications is a group of software for LLRF system in order to:

Optimize operating parameters of LLRF controller

System Diagnostics

Assist operator to simplify operation

4

Examples for applications

DAC DC offset calibration for LLRF controllerGoal: Assist operator with zero power calibration

Cavity quench detectionGoal: Diagnostic cavity limitation

Adaptive feed forwardGoal: Optimize controller’s feed forward parameters

5

DAC DC Offset Calibration-- Assist operator with zero power calibration

6

DAC DC offset calibration-- Introduction

DAC

Vector modulator PA1 PA2 Klystronininin jQIV +=

outoutout jQIV +=

RF Gate

Pick 1 Pick 2 Pick 3 Pick 4

DC offset will cause problem to set zero power (gradient)

Unknown power offsets will cause error for applications that take the RF decay data (assuming there is no power at the end of the RF pulse)

Zero power is obtained by removing the offset at DAC signals (currently by hand)

DC offset changes with time, so calibration has to be done from time to time

7

DAC DC offset calibration-- Driving chain error model

Iin

Qin

Ioffset

Qoffset

Vector rotation,

multiplied by

(Ki+j*Kq)

I'in

Q'in

cos(ωt)

gsin(ωt+φ)

Demodulation (by RF signal measurement)

Pick n

Iout

Qout

Unknown parameters: Known parameters:Calibration strategy: Linear fittingAssumption: RF signal measurement is perfect; System is linear

ϕ,,,, gjKKQI qioffsetoffset +ω,,,, outoutinin QIQI

8

ResultsEquation for input/output

⎩⎨⎧

++=++=

ldQcIQkbQaII

ininout

ininout

( )

DAC DC offset calibration-- Formulas used

( )( ) ( )

⎪⎪⎪

⎩

⎪⎪⎪

⎨

⎧

+=+=

==

==

==

+−=+=

nmlfek

dQncIm

bQfaIe

gKdgKc

gKKbgKKa

offsetoffset

offsetoffset

iq

iqqi

,

,

,

cos,cos

sin,sin

ϕϕ

ϕϕ

⎪⎩

⎪⎨⎧

+=

+=

offsetoffset

offsetoffset

dQcIl

bQaIk

( )

( )⎪⎪⎪⎪⎪⎪

⎩

⎪⎪⎪⎪⎪⎪

⎨

⎧

+=

=

=

⎟⎠⎞

⎜⎝⎛ −

⎟⎟⎠

⎞⎜⎜⎝

⎛+

=

i

q

q

iq

i

KKb

g

Kcg

dcKK

dcba

dc

K

ϕ

ϕ

sin

cos

1

1

2

2

At least 3 pairs of input/output points are needed to perform the linear fitting.

9

DAC DC offset calibration-- Test results

Gradient / MV/m Ioffset Qoffset Ki Kq g φ / deg

3 points 1 4952 10525 -6. 737 e-6 1. 306 e-6 0.9976 -0.77

8 points 1 4875 10505 -6.744 e-6 1.298 e-6 0.9967 -0.16

16 points 5 4937 10255 -6.678 e-6 1.307 e-6 0.9988 -0.16

32 points 5 4936 10276 -6.708 e-6 1.245 e-6 0.9987 -0.11

64 points 1 4847 10467 -6.747 e-6 1.270 e-6 0.9963 -0.01

Measured Values

10

DAC DC offset calibration-- Test results

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

x 104

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

x 104IQ input from the feed forward table, set point gradient = 5 MV/m

I / arbitrary unit

Q /

arbi

trary

uni

t

Input by feed forward signal:* Point Number: 32* Gradient Set Point: 5 MV/m

-0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2

-0.05

0

0.05

0.1

0.15

0.2

IQ output from the klystron forward signal measurement, set point gradient = 5 MV/m

I / arbitrary unit

Q /

arbi

trary

uni

t

Output from klystron RF signal:* Point Number: 32* Gradient Set Point: 5 MV/m

0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08

0.04

0.05

0.06

0.07

0.08

0.09

IQ output from the klystron forward signal measurement, set point gradient = 1 MV/m

I / arbitrary unit

Q /

arbi

trary

uni

t

Output from klystron RF signal:* Point Number: 8* Gradient Set Point: 1 MV/m

-5000 -4000 -3000 -2000 -1000 0 1000 2000 3000 4000 5000

-4000

-3000

-2000

-1000

0

1000

2000

3000

4000

IQ input from the feed forward table, set point gradient = 1 MV/m

I / arbitrary unit

Q /

arbi

trary

uni

t

Input by feed forward signal:* Point Number: 8* Gradient Set Point: 1 MV/m

11

DAC DC offset calibration-- Summary

The current algorithm need to interrupt the normal operation because it has to change the gradient set point and rotate the feed forward signal

When RF system first start up, offset calibration can be done with klystron off, so that we can avoid klystron interlock trip by unexpected peak driving power

In the future, the calibration can be done without interrupting normal operation by introducing a small calibration pulse after the main RF pulse (see the figure below)

0 200 400 600 800 1000 1200 1400 1600 1800 20000

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5x 104

Time / μs

Am

plitu

de /

arbi

trary

uni

t

Calibration pulse after the normal RF pulse by feed forward table

Feed forward signal for normal RF pulse

Calibration pulse

12

Cavity Quench Detection-- Diagnostic cavity limitation

13

Cavity quench detection-- Problem description

Cavity quench can cause unstable RF field or even beam loss, and increase the cryo heat load

Goals for this application: Detect quench for each cavity to inform operators and cryogenics

0 200 400 600 800 1000 1200 1400 1600 1800 20000

5

10

15

20

25

Time / μs

Gra

dien

t / a

rbitr

ary

unit

Cavity gradient for ACC1, cavity No2 quench

Cav1Cav2Cav3Cav4Cav5Cav6Cav7Cav8

500 600 700 800 900 1000 1100 1200 1300 1400 1500

10

11

12

13

14

15

Time / μs

Cavity gradient for ACC1, cavity No2 quench (zoom)

Gra

dien

t / a

rbitr

ary

unit

Cav1Cav2Cav3Cav4Cav5Cav6Cav7Cav8

14

Cavity quench detection-- Solution

Existing solutions at FLASH:At the DSP system, quench is detected by monitoring the change of the vector

sum pulse shape (work only for vector sum; can not distinguish the pulse shape change by detuning or beam loading)

Measure loaded Q of each cavity at the RF pulse decay part (by Valeri Ayvazyan. pulse to pulse quench detection; works for each cavity; precise)

Solution proposed here: Measure the loaded Q of each cavity during the RF pulse (real time intra-pulse quench detection; precise)

If the loaded Q drops larger than the threshold, quench event will be generated

Principle: the cavity equation is used for loaded Q measurement

( )

0

0

2/12/12/1 2

ZQrC

IRVCVjdt

dVbLforc

c

ω

ωωωω

⎟⎟⎠

⎞⎜⎜⎝

⎛=

+′=Δ−+

15

Cavity quench detection-- Test results

2 4 6 8 10 12 142.4

2.6

2.8

3

3.2x 10

6

Load

ed Q

of c

avity

No1

2 4 6 8 10 12 142.4

2.6

2.8

3

3.2x 10

6

Load

ed Q

of c

avity

No2

2 4 6 8 10 12 142.4

2.6

2.8

3

3.2x 10

6

Load

ed Q

of c

avity

No3

2 4 6 8 10 12 142.4

2.6

2.8

3

3.2x 10

6

Load

ed Q

of c

avity

No4

2 4 6 8 10 12 142.4

2.6

2.8

3

3.2x 106

Load

ed Q

of c

avity

No5

2 4 6 8 10 12 142.4

2.6

2.8

3

3.2x 106

Load

ed Q

of c

avity

No6

2 4 6 8 10 12 142.4

2.6

2.8

3

3.2x 106

Load

ed Q

of c

avity

No7

2 4 6 8 10 12 142.4

2.6

2.8

3

3.2x 106

Load

ed Q

of c

avity

No8

Loaded Q measurement at the RF decay part for each cavity of ACC1, the x number means 14 times measurement with different set pointgradient (from 9.3MV/m to 10.6MV/m, 0.1MV/m as increment steps)

16

Cavity quench detection-- Test results

Loaded Q measurement during RF flattop for each cavity of ACC1, the curves for each cavity means 14 times measurement with different set

point gradient (from 9.3MV/m to 10.6MV/m, 0.1MV/m as increment steps)

200 400 600 8002

2.2

2.4

2.6

2.8

3

3.2x 106

Flattop time / μs

Load

ed Q

of c

avity

No1

200 400 600 8002

2.2

2.4

2.6

2.8

3

3.2x 106

Flattop time / μs

Load

ed Q

of c

avity

No2

200 400 600 8002

2.2

2.4

2.6

2.8

3

3.2x 106

Flattop time / μs

Load

ed Q

of c

avity

No3

200 400 600 8002

2.2

2.4

2.6

2.8

3

3.2x 106

Flattop time / μs

Load

ed Q

of c

avity

No4

200 400 600 8002

2.2

2.4

2.6

2.8

3

3.2x 106

Flattop time / μs

Load

ed Q

of c

avity

No5

200 400 600 8002

2.2

2.4

2.6

2.8

3

3.2x 106

Flattop time / μs

Load

ed Q

of c

avity

No6

200 400 600 8002

2.2

2.4

2.6

2.8

3

3.2x 106

Flattop time / μs

Load

ed Q

of c

avity

No7

200 400 600 8002

2.2

2.4

2.6

2.8

3

3.2x 106

Flattop time / μs

Load

ed Q

of c

avity

No8

This method also works in presence of beam

17

Cavity quench detection-- Test results

Loaded Q measurement of cavity No.2 at ACC1 during the RF pulse with different set point gradient

200 400 600 800 1000 1200 1400 1600 1800 2000 2200 24002

2.2

2.4

2.6

2.8

3

3.2x 106

Tme / μs

Load

ed Q

The loaded Q measurement for the whole RF pulse

SP = 9.3 MV/mSP = 9.4 MV/mSP = 9.5 MV/mSP = 9.6 MV/mSP = 9.7 MV/mSP = 9.8 MV/mSP = 9.9 MV/mSP = 10.0 MV/mSP = 10.1 MV/mSP = 10.2 MV/mSP = 10.3 MV/mSP = 10.4 MV/mSP = 10.5 MV/mSP = 10.6 MV/m

18

Cavity quench detection-- Summary

The algorithms for the intra-pulse real time quench detection is evaluated, which is currently developed by matlab, and not yet available for operators

Quench detection at RF decay is easy, which can be implemented at LLRF ATCA CPU

Quench detection during RF pulse in real time is relatively difficult to realize, which must be implemented at faster processors such as DSP or FPGA

Whether or not implement the real time quench detection strongly depends on the requirements from the operation group!

19

Adaptive Feed Forward-- Optimize controller’s feed forward parameters

20

Adaptive feed forward-- Problem description

Goals of the application:

Compensate the repeating errors of the system

When system status changes, such as the gradient or beam condition change, adapt the feed forward table for new working point setting

0 200 400 600 800 1000 1200 1400 1600 1800 20000

2

4

6

8

10

12

14

RF System(Including klystron and

cavities)0 200 400 600 800 1000 1200 1400 1600 1800 2000

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

200 400 600 800 1000 1200 1400 1600 1800 20000

0.5

1

1.5

2

2.5

3x 107

21

Adaptive feed forward-- Solutions

Time inversed filter based solutionBy Alexander Brandt

The most robust and mature one working for FLASH

Easy to implement

Inversed black box model based solution (learning feed forward)By Christian Schmidt

The most precise one

The black box model parameters for the RF system have no obvious physical meaning

Inversed gray box model based solutionThe topic here

The gray box model parameters for the RF system have obvious physical meaning,

which will be helpful to understand the system

It is possible to predict the required beam compensation signal for the desired beam

setting and compensate the beam loading in advance

22

Adaptive feed forward-- Gray box model, open loop

Vector sum set point

Feedback control

Feed forward

DAC

Vector modulator PA1 PA2 Klystron

… … (at most 32

cavities)

… … LO

ADC

ADC

… …

Vector sum calibration

… … … …

Vector sumMeasurement chain loop phase Filter

Down converter

Demodulation

Demodulation

A

C

B

A

Vdac

Vsum

D

Vset Vkly

Vff

A to B: modeled as a time varying gain and phase shift, marked as complex gain G

B to C: modeled as an effective single cavity, so the open loop model is

( )0

02/12/1 ZQ

rMGVCMVjdt

dVdacklysum

sum ωωωω ⎟⎟⎠

⎞⎜⎜⎝

⎛==Δ−+ ，

23

Adaptive feed forward-- Gray box model, closed loop

Vector sum set point

Taking into account the feedback (here is a P controller), the closed loop gray box model is

Feedback control

Feed forward

DAC

Vector modulator PA1 PA2 Klystron

… … (at most 32

cavities)

… … LO

ADC

ADC

… …

Vector sum calibration

… … … …

Vector sumMeasurement chain loop phase Filter

Down converter

Demodulation

Demodulation

A

C

B

A

Vdac

Vsum

D

Vset Vkly

Vff

( )0

02/12/12/1 ZQ

rMGVCMVjGPCMdt

dVffklysumkly

sum ωωωωω ⎟⎟⎠

⎞⎜⎜⎝

⎛==Δ−++ ，

24

Adaptive feed forward-- Test results, the identified model elements

0 200 400 600 800 1000 1200 1400 1600 1800 20002

2.2

2.4

2.6

2.8

3

3.2

3.4

3.6

3.8

4x 106

Time / μs

Load

ed Q

Loaded Q of vector sum effective cavity during the RF pulse

0 200 400 600 800 1000 1200 1400 1600 1800 2000-200

-100

0

100

200

300

400

Time / μs

Det

unin

g / H

z

Detuning of vector sum effective cavity during the RF pulse

0 200 400 600 800 1000 1200 1400 1600 1800 20004

5

6

7

8

9

10

11

12x 10-5

Time / μs

Gai

n / a

rbitr

ary

unit

Gain from the DAC output to the klystron output

0 200 400 600 800 1000 1200 1400 1600 1800 2000-30

-25

-20

-15

-10

-5

0

5

10

Time / μs

Pha

se s

hift

/ deg

Phase shift from the DAC output to the klystron output

25

Adaptive feed forward-- Test results, closed loop AFF

The adaption is from the start of the flattop

Transient can be removed by starting the adaption a little earlier than the flattop start point

26

Adaptive feed forward-- Test results, convergence and stability

Adaption gain =

0.1

27

Summary

LLRF applications is part of LLRF system for better performance, system diagnostic and easy operation supportingMost applications are based on the RF cavity physical modelApplication development will be on SIMCON DSP development system, but the computation resource is quite limited to implement all necessary applications. Applications will be immigrate to the ATCA system when it is ready.

28

Thank you!

Progress on LLRF applications development OutlineGoals of LLRF applicationsExamples for applicationsDAC DC Offset Calibration�-- Assist operator with zero power calibration �DAC DC offset calibration�-- IntroductionDAC DC offset calibration�-- Driving chain error modelDAC DC offset calibration�-- Formulas usedDAC DC offset calibration�-- Test resultsDAC DC offset calibration�-- Test resultsDAC DC offset calibration�-- SummaryCavity Quench Detection�-- Diagnostic cavity limitationCavity quench detection�-- Problem descriptionCavity quench detection�-- SolutionCavity quench detection�-- Test resultsCavity quench detection�-- Test resultsCavity quench detection�-- Test resultsCavity quench detection�-- SummaryAdaptive Feed Forward�-- Optimize controller’s feed forward parametersAdaptive feed forward�-- Problem descriptionAdaptive feed forward�-- SolutionsAdaptive feed forward�-- Gray box model, open loopAdaptive feed forward�-- Gray box model, closed loopAdaptive feed forward�-- Test results, the identified model elementsAdaptive feed forward�-- Test results, closed loop AFFAdaptive feed forward�-- Test results, convergence and stabilitySummaryThank you!