LLRF implementation for the FLASH 9mA
run in September
John Carwardine
9 February 2009
Introduction
• Problem statement
– Resource limitations put in jeopardy our previous plans to
have a SimconDSP system at ACC456 for the 9mA demo
• To be discussed today
– Could the existing LLRF systems support a full 9mA demo
in Sept? (DSP systems at ACC23 &ACC456)
– Develop a prioritized list of additional functionality that
would improve the confidence level for a successful test
– Resources
• Note: we originally pursued the SimconDSP path to take
advantage of synergy with work that has to be done for the new
ATCA system
Goals of the September 9mA studies
1. Run 2400 bunches, 3nC/bunch @ 3MHz (9mA, 800us)
2. Run at gradients close to quench on ACC456
3. Demonstrate ILC-like energy stability over long bunch trains and
many pulses (+/-0.1%, +/-0.1deg)
4. Measure the required RF power overhead on ACC456
5. Exception handling, graceful response and recovery
6. Stable and repeatable operation over 1-2 week period
•
These are difficult goals, eg, we don’t know how the LLRF systems
will behave with high beam loading
•
•
If we had to, which goal would we give up first…?
How do we define a successful test?
Desired LLRF baseline for the 9mA experiment
(as discussed previously)
•
•
•
ACC456
– 3x SimconDSP system with new high IF boards (fwd/ref pwr not needed)
– Do not attempt to develop a 9x SimconDSP system – instead, put the
resources into making a SimconDSP system for ACC23.
ACC23
– 2x SimconDSP system with new high IF boards (fwd/ref pwr not needed)
– Use the 3-week shutdown to temporarily hook up the SimconDSP system
– Needs to also be upgraded in order to get best overall performance and
to have consistent algorithms and behavior.
Advantages of SimconDSP with high IF for the 9mA Expt.
– Lower noise floor due to higher digitization rate
– Higher IF gives higher open-loop bandwidth
– There are spare resources on the board to allow implementation of
additional key functions over what can be done with the DSP systems
Desired LLRF baseline for the 9mA experiment
(as discussed previously)
•
•
•
ACC456
– 3x SimconDSP system with new high IF boards (fwd/ref pwr not needed)
– Do not attempt to develop a 9x SimconDSP system – instead, put the
resources into making a SimconDSP system for ACC23.
ACC23
– 2x SimconDSP system with new high IF boards (fwd/ref pwr not needed)
– Use the 3-week shutdown to temporarily hook up the SimconDSP system
– Needs to also be upgraded in order to get best overall performance and
to have consistent algorithms and behavior.
Advantages of SimconDSP with high IF for the 9mA Expt.
– Lower noise floor due to higher digitization rate
– Higher IF gives higher open-loop bandwidth
– There are spare resources on the board to allow implementation of
additional key functions over what can be done with the DSP systems
Comparing SimconDSP and DSP for ACC23 and ACC456
DSP systems (presently installed in ACC23 and ACC456)
• Existing, proven, feedback gain limitations, no capacity for extra features
– Some needed functions have already been implemented on the DSP
system, eg manual beam loading compensation
– There is a lot of operational experience using the DSP systems to control
ACC23 and ACC456, and the installations are well understood
– We used the DSP systems to run 550 bunches with 3mA beam loading
SimconDSP (presently installed in ACC1, only limited testing at ACC456)
• Not existing, unproven, promises better performance, has spare capacity
– Higher IF is possible. Even with 1MHz IF, higher sampling rate (50MS/s)
gives better noise floor and eliminates 250kHz ripple
– Should be possible to get higher feedback gain than DSP systems
– Has spare capacity for additional processing (not true for DSP systems)
Work involved in providing SimconDSP systems
•
•
•
•
•
Boards have to be built and tested
FPGA code development
DOOCS server software
Installation and testing in FLASH
….
• Resources would have to be found or re-prioritized for this to be
done before September.
• Worth considering…?
– Might be possible to implement the Fermilab ISICon firmware
with the SimconDSP boards
Summary (1)
• A proven SimconDSP system would likely increase the chances of
success and should give better energy stability
• The existing DSP systems should be adequate to meet the most
important goal of the 9mA studies, ie 2400 bunches, 9mA
• Controlling beam loss will be key to success. This means better
regulation of beam energy…
Desired LLRF capabilities for 9mA Experiment
Capability
ACC456
ACC23
High feedback gain (>>40). Feedback gain is ramped up during fill time to reduce RF
transient peak power. Smoothing algorithm on feed-forward tables
Yes
Yes
Pre-load beam loading compensation based on nominal bunch pattern and charge
provided by DOOCS
Yes
Yes
MPS: respond to trips, send quench signal, graceful response and recovery
Yes
Yes
Simple cavity quench detection - detect at end of the RF pulse, inhibit the next pulse
(eg integral of cavity probe, loaded Q at end of pulse) – implemented in DOOCS
Yes
?
Eliminate 250kHz ripple. Add 2-sample boxcar averager to I&Q calculation in order to
remove DC offset
Yes
Yes
More accurate and faster calibration of vector sums. Vector sum from forward &
reflected power measurements?
Yes
Yes
ACC5+6
No
Online detuning calculation at 5Hz – implemented in DOOCS
Yes
Yes
Create new DOOCS panels for frequency and tuners
Yes
Yes
Fix the cross-talk problem which was observed on high power chain during filling
time at ACC23
Yes
Yes
Lorentz-force detuning compensation using piezos
4 Feb 2009 (JAC)
Higher feedback gains
• Benefit to 9mA studies
– Keep beam losses low by reducing energy excursions
– Minimize energy excursions during long pulses and from pulse to pulse
– Avoids the need for slow adaptive feed forward
• Achieved:
– We were able to get a useful gain of >70 on DSP systems during
January studies. SimconDSP systems promise even higher gains
– Is a gain of 70 enough?
In good
shape?
In progress
• Related tasks:
– Implement gain ramping to minimize transient peak RF power
– Implement better ways to generate feed-forward tables
Beam loading compensation
• Beam-loading compensation term added to feed-forward tables
based on beam current and pulse length
– Manual compensation already available on DSP systems
OK
To do – Automate setting of compensation term through DOOCS
– Add a time-varying setpoint term (slope) to compensate for slope
To do
in beam current during long flat-top
• Benefit
– Reduce likelihood of mistakes in setting up the beam loading
compensation.
– Make it easier to operate the system off-hours when experts
aren't around
– Help compensate beam loading effects if gain (~70-80) appears
to be not high enough with full beam loading
High priority
Exception handling: MPS
• Minimal
– LLRF receives MPS enable/inhibit.
– Upon inhibit, terminates RF pulse
– Benefit
High
priority
• Necessary to avoid large increase in RF voltage
• Preferred
– Graceful response to MPS and graceful recovery
• We need to decide specifically what that means
– Benefit
• A component of demonstrating smooth operation
Medium
priority
Quench detection
• Detect quenches, turn of beam and RF
• Benefit
– First: prevent hard quenches and need for cryo recovery
– Second: demonstrate graceful response/recovery
• ‘Simple’ detection
– Detect after pulse from cavity field waveforms
– Implement in DOOCS or Matlab, inhibit next pulse via
MPS
High
priority
Eliminate 250kHz ripple
• Comes from DC offsets in I&Q sampling signal chain
• Not observed in Simcon systems due to faster digitization rate.
• Impact:
– Eats into energy stability budget and can increase beam loss
• Currently compensated using ripple correction tables set by an
algorithm implemented in Matlab. Algorithm must be run periodically
– Is this adequate?
• A 2-point boxcar averager on the I&Q data streams
in
To is
doproposed
or not to do…?
order to cancel the DC term
More accurate & faster calibration of vector sums
• Concern
– Calibration takes a long time and requires expert care
– Calibrations must apparently be done frequently
• Problem statement:
– Can we do better?
• Subject of separate meeting
– Review present algorithms and limitations
– Explore improvements and/or new algorithms
– Explore ways to automate the process
High priority
Automated/rapid cavity tuning
• Rapid cavity tuning, eg when changing gradients
• Current system requires careful manual tuning by Valeri
and it is time-consuming to do
• We would like something that is more automated and
doesn't require Valeri to get 'decent' results
• Benefit
– Reduce the time consumed in re-tuning the cavities when
we change operating conditions
Requires study and
work to implement
Medium-High
priority
Lorentz force detuning compensation
In good
• Use piezos to compensate for LFD
shape
– Has already been tested in bench-top experiments, currently
being implemented as permanent hook-up.
• Benefit
– Reduce peak RF power needs (important when running at the
limit of gradient and RF power)
• Provide real-time measurement of de-tuning during flat-top
– Algorithm should be implemented in DOOCS.
Medium
priority
Other
• Fix cross-talk observed in the vector modulator
• Make additional DOOCS panels, other DOOCS server
changes,…
Desired LLRF capabilities for 9mA Experiment
Capability
ACC456
ACC23
High feedback gain (>>40). Feedback gain is ramped up during fill time to reduce RF
transient peak power. Smoothing algorithm on feed-forward tables
Yes High Yes
Pre-load beam loading compensation based on nominal bunch pattern and charge
provided by DOOCS
Yes High Yes
MPS: respond to trips, send quench signal, graceful response and recovery
Yes
Simple cavity quench detection - detect at end of the RF pulse, inhibit the next pulse
(eg integral of cavity probe, loaded Q at end of pulse) – implemented in DOOCS
Yes High ?
Eliminate 250kHz ripple. Add 2-sample boxcar averager to I&Q calculation in order to
remove DC offset
Yes High Yes
More accurate and faster calibration of vector sums. Vector sum from forward &
reflected power measurements?
Yes High Yes
Lorentz-force detuning compensation using piezos
Yes
ACC5+6
No
Online detuning calculation at 5Hz – implemented in DOOCS
Yes
Yes
Create new DOOCS panels for frequency and tuners
Yes
Yes
Fix the cross-talk problem which was observed on high power chain during filling
time at ACC23
Yes
Yes
SimconDSPs for ACC456 and ACC23
??
4 Feb 2009 (JAC)
Summary (2)
• There are no obvious show-stoppers if we use the
existing DSP systems in ACC23 and ACC456
– but SimconDSPs would provide better performance in the
long run, and they could be done in parallel with DSP
• There are high priority tasks regardless of the hardware:
– Higher feedback gain, beam loading compensation,
interfaces to/from MPS, quench detection, calibrations!
• The less progress that is made, the greater the risk of
not accomplishing the goals.
• So…what resources can we plan for?