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!