Wednesday
May 17th, 14:30
CCC
conference room 874/1-011
There were no comments on the minutes of the previous meeting.
The LHCCWG team welcomed three new members, namely Reyes Alemany Fernandez, Laurette Ponce, and Walter Venturini Delsolaro. All three will soon start as EICs.
In response to
an earlier request from the LTC, Roger has drafted a summary of our discussions
on an LHC operation period with 75-ns spacing, referring to the presentation
by Gianluigi and Frank at an earlier LHCCWG
meeting. Roger’s draft
note was circulated prior to today’s meeting. Several LHCCWG members have
sent further comments, which have been incorporated. Roger will
present the final assessment and, in particular, our recommendation that a
75-ns period be kept, at the next LTC on May 24.
Roger reported
that the special meeting which had been organized by Steve on the interference
of hardware and beam commissioning for a possible sector test in 2007
considered three different installation scenarios. The ‘nominal’ scenario would
see the first beam in spring 2008. The two alternative scenarios accelerate the
installation. However, even for the most
optimistic proposal presented, collisions in 2007 are excluded. No final
conclusion was reached, and a follow-up meeting will be held next Tuesday.
Roger highlighted that if collisions are postponed to 2008 a sector test
in 2007 would be back on the horizon.
Roger presented an LHCCWG schedule, listing the topics, speakers and days of past and future meetings.
Andy reviewed rf
issues for beam intensities up to 43 bunches with 4e10
protons per bunch, corresponding to 3 mA beam
current. He distinguished three sub-systems, namely the high-power rf, the low-level rf plus beam
control, and the synchronization.
First, for the high power rf
system, beam instability due to loss of control of a cavity and the associated
machine protection issue become important only for total beam currents above
250 mA, i.e., above about half the nominal. However,
the protection of cavities and loads
against power coupled out from the beam may
require beam interlocks already for much lower beam currents, i.e., higher
than 0.6 mA.
Second, the low-level rf
system contains three types of feedback loops: (1) a tuner loop which adjusts
the cavity resonant frequency so as to minimize the klystron current, (2) an rf feedback loop, which reduces the loaded quality factor Q
of the cavity by a factor 20 from 20000 at injection and by a factor 180 from
180000 at top energy, which compensates for transient beam loading and which
ensures longitudinal stability, and (3) a 1-turn delay feedback, which adds
another factor of 10 impedance reduction on the revolution frequency sidebands.
The first two feedback loops will be available and commissioned from day 1. By
contrast, the 1-turn delay feedback will not be commissioned at the start, for
two reasons: this system is rather complex and it is not needed initially. The
intensity in 43-bunch operation is a factor 150 below the projected
coupled-bunch instability threshold without the 1-turn delay loop. On the other
hand, the tuner loop will be used from the beginning to realize a “half detuning” technique, by which
klystron-power transients are minimized. The half detuning method was not
employed at LEP. This technique requires
information on the intensity of the injected beam prior to the actual
injection, provided by the timing system.
Andy mentioned that the cavity fill time is of
the order of 10 microseconds, but that the tuner loop response time is much
faster than this. In first order, the position and number of bunches are not
crucial, and only the average bunch intensity (or the local average beam
current over a short section of the ring) needs to be known for adjusting the
feedback. Responding to a question by Paul, Andy explained that the couplers
are tuned automatically. Likely a software command will be sent somewhere
between injection and top energy to change the coupler setting and to increase
the loaded Q.
Four different beam control loops exist: (1) a
phase loop, which locks the rf phase onto the beam
and which can damp injection oscillations of a single batch, (2) a synchro loop which, more slowly, locks the phase of RF and
beam onto a frequency programme, (3) a radial loop
which maintains the horizontal beam position at a radial pick up, and (4) a
longitudinal feedback which damps injection oscillations and coupled-bunch
modes, and which should ideally be commissioned before starting with multiple
injections. The phase, synchro and radial loops must be commissioned with
a single pilot bunch.
Stephane asked for the potential coupling between the
two rings via their respective radial loops. Andy replied that at injection
both ring rf systems are locked to the SPS rf. After injection the two rf
systems can be unlocked and one or both radial loops be used if desired. The
response rate of the phase and radial loops is of the order of the synchrotron
frequency.
Jorg and Jean-Pierre pointed out that a single
radial pick up could make the radial loop respond to betatron
motion or noise, and that in RHIC the radial loop now acts only on the momentum
component of the horizontal orbit motion which is extracted from two pick ups
with great success. Ralph S. remarked that in a similar fashion one could
process the readings of all arc BPMs for an improved
performance. Gianluigi mentioned that the radial loop
is not used for normal SPS operation, since the SPS cycle is very reproducible
without loop. A general consensus emerged that the LHC radial loop will likely
be used only for the commissioning, and that it will afterwards be replaced by
a frequency programme.
Stephane expressed some concern about the expected
level of reproducibility in the LHC. Jorg quoted a
circumference variation larger than 1 mm due to the tides, which could in
principle be corrected by feedforward. Ralph A. asked
whether it is planned to ramp with the radial loop active. The following
discussion suggested that this would not be a problem as the loop response is
slow. Stephane mentioned that with 2 units change in
the main dipole field, variations of up to +/-10 Hz may be expected from cycle
to cycle. Oliver pointed out that these numbers are consistent with HERA
experience, where it was rather easy not to capture the injected beam, without
careful adjustments.
Third and lastly, Andy addressed the rf synchronization system, which organizes the bunch into
bucket transfer from the SPS to the LHC, and generates pre-pulses for the
extraction and injection kickers. This synchronization
requires injection bucket number and ring identifier which must be supplied by
the timing system (in addition to the beam intensity needed by the tuner
loop).
Jan asked whether in case of a beam dump
initiated by the rf system, the klystron power would be
reduced immediately. Andy replied that the klystron power is reduced when the
beam permit disappears, e.g., when the machine protection system requests a
beam dump. Jorg conceived a more complicated scenario
where the beam permit goes down but the beam dump does not actually fire. In
this case reducing the klystron power will lead to debunching.
According to Ralph A., this scenario is not a serious problem as one could dump
the debunched beam at the expense of a quench.
Paul asked which longitudinal diagnostics will
be available. Andy answered that the main diagnostics is a resistive-wall
monitor “wideband pick up”, which
feeds a mountain-range display. Oliver recalled a past discussion in the LTC,
according to which it was unclear whether this monitor signal would be
available in the control room. Andy’s reassuring reply was that the
mountain-range signal will certainly be available
in the control room through OASIS.
Gianluigi remarked that the phase or radial loop
signals can also be made available via OASIS.
Rudiger clarified that even though injections into the
wrong bucket are possible, a separate protection
system ensures that an injection will never occur into a bucket of the abort
gap.
Oliver asked who will check that the bucket
number is ok. The correctness of the
bucket numbers needs to be verified prior to the ramp. Either the
longitudinal bunch profiles or a fast BCT (Jorg)
could be used for such checks. Paul commented that this problem is being
addressed by an OP project and that a person has been put in charge of this
issue. Rhodri pointed out that the underlying software for either signal is
OASIS, but that higher level software will be needed to e.g., extract and
display the bunch patterns.
Stephane asked whether selected bunches could be
excited longitudinally. Andy’s reply was that the phase loop cannot perform
bunch-by-bunch excitation. Oliver recalled a discussion with Trevor according
to which a maximum excitation by 10 degrees would be possible at 7 TeV via rf phase modulation, and
that phase modulation is much faster than frequency modulation. Gianluigi added
that a longitudinal noise excitation of all bunches is anyway foreseen during
the ramp, and Andy explained that the latter can be realized by injecting noise
at 2Qs, using the longitudinal feedback.
Ralph emphasized that a clear prediction for the momentum reproducibility
from fill to fill, and for the movements in the radial position, is highly
desirable as this would represent an important input for the momentum cleaning
studies. As a related topic he mentioned the interference of chromaticity
measurements with beam cleaning. ACTION:
Ralph A.
In his presentation
For
43-bunch operation the tolerance on the
orbit stability appears quite relaxed, about 1-1.5 sigma, which is to be compared with 0.3
sigma for nominal intensity. Ralph S. emphasized though that the orbit
is not a “play-parameter” for optimization. In order to guarantee the proper functioning of the collimation and
machine protection system, it has to be insured that the primary and secondary
collimator always define the global primary and secondary aperture
respectively. This requirement may be compromised by for example orbit bumps in
the arc which may lead to magnet quenches, or, in combination with a fast
failure, even to damage of the machine (e.g. orbit bump and tune/aperture
kicker). In order to test the orbit with respect to the actual available
aperture, an automated aperture check procedure has been proposed (see slides
for details). Ralph pointed out that the distinction between local and
global tolerances is not that meaningful, as the tolerances are similarly tight
at many locations around the machine. The required initial momentum stability
at injection from the SPS is 1e-4. Fast dynamic orbit excursions are expected
during the snapback and even larger ones during the squeeze. In case of the squeeze, these excursions
may reach up to 30 mm if not corrected,
assuming a random worst-case quadrupole misalignment
of 0.5 mm r.m.s. in the LHC.
The squeeze was considered as less critical, since one could slow it down or divide it into more and smaller steps, which is less obvious for the snapback. Details on the predicted dynamic perturbations can be found in the slides.
Ralph
next reviewed the common orbit correction strategy which is to first establish
a reference and then to stabilize this reference using a feedback. The feedback
correction is split into a
space and a time component. In space, the SVD algorithm is used, which
minimizes both orbit and corrector strengths. The number of singular values
retained in the pseudo matrix inversion balances robustness versus precision. In the time domain, a PID controller extended
by a Smith predictor is employed [O.J.M. Smith, UC Berkeley, 1957]. The Smith
predictor compensates for the effect of dead time and may be used for nominal operation.
The LHC orbit feedback algorithm was tested in numerous simulations. It works well even with 20% of BPMs failing.
Ralph
A. commented that if several successive BPMs are bad,
the orbit in an entire region would not be known. Ralph S. responded that successive BPMs are processed in different front ends to minimize the
probability of adjacent BPMs failing simultaneously.
In principle an interlock could be added to dump the beam if BPMs in an extended region fail. (Except for dedicated BPMs in IR6, the present
LHC baseline does not foresee a software-interlock on the global orbit
position.)
Tests
of the LHC BPMs in the SPS revealed two common types of BPM failure:
missing orbit information or spikes. Taking this into account, the feedback
response to failed BPMs is different on a short and
long time scale. On a short time scale the BPM reading is replaced by a zero
order holder, which assumes that the orbit stays the same as for the last
reading. If the BPM does not recover over a longer period, a new matrix without
this BPM is computed and used for future feedback activity. For detecting
failed BPMs, cuts are applied in space and time
domain. In space domain, the monitor
is rejected if its reading exceeds three times the value of the reference orbit
r.m.s. In time domain the cut is typically applied
also at three sigma, where sigma now refers to the
temporal orbit r.m.s. noise during the last 'n'
seconds measured with the BPM. It is difficult though to detect very slow or
systematic drifts in the BPM readings. The feedback matrices are usually based
on the theoretical optics, but it is also possible to use
measured orbit response matrices.
They are re-calculated on-the-fly for each phase of the cycle, whenever
the optics, BPM or COD configuration changes.
Ralph
A. asked what the feedback response would be in case of a fast (~ms scale)
orbit change and whether in such case all BPMs of the
feedback would be disabled. Jorg commented that even
the snapback effect if fairly slow, with mm orbit changes over many seconds. Thijs commented on the voltage limitations of the
correctors. Ralph S. showed how the corrector rate limits are taken into
account in the feedback algorithm by adjusting the ramp rate of all correctors
with a common factor. Stephane confirmed that the snapback
is the fastest change expected. He asked how the correction is sent. Jorg’s answer was in a number of steps according to the
feedback gain chosen. Ralph S. pointed out that the feedback has a lot of
safety built in from oversampling (BPM spacing at 45
phase advance). Paul asked for the cross
talk between the two rings. Ralph S. replied that the cross talk is always
taken into account. If operation
permits, it is favourable to not use common
correctors when only one beam is present, since perfect corrections for one
beam may potentially have a deteriorating effect on the other if not measured
with both beams at the same time. It is preferred to establish circulating
beams in both rings prior to injecting intermediate or higher intensities, if
two-beam operation is foreseen during that fill. Oliver commented that
in case the correctors compensate for magnet misalignments, the same corrector
setting would be correct for either beam. At top energy misalignments might be
a dominant error source. Ralph A. mentioned that there is additional cross talk
in the orbits of the two rings due to long-range beam-beam collisions. JPK
pointed out the importance of correction strategy for the low-beta region,
where phase advances are tiny.
Ralph S. showed that a single undetected BPM step failure (i.e., below 3 sigma in time-variation rms (e.g., rms~60 micron) or orbit rms (e.g., rms~300 micron), respectively) changes the orbit locally by between 0.01 and 0.4 times the BPM error step. Adding the measured white noise at all BPMs (<20 micron per orbit measurement of a single pilot bunch) leads to an error of 0.001 times the beam sigma at injection and 0.02 times the beam sigma in collision.
Rudiger asked how the dipole correctors are ramped and, together with Paul, whether they are scaled with beam energy. Jorg commented that the correction of the reproducible part of the orbit change, e.g., the one recurring from cycle to cycle, will not be left to the orbit feedback but after a few cycles be taken care of by feedforward. Paul remarked that the SVD corrections spreads the errors around the machine. Jorg replied that the bare orbits and errors can be reconstructed from the feedback data (BPM readings and corrector settings). Ralph S. quoted an example of 5-10 GBytes of orbit logging data alone, for each fill.
Corrector (COD) failures are expected at a rate of 1 every 5 days with an expected MTBF of 1e5 hours. The maximum corrector deflection angle is 1.3 mrad at injection and 0.083 mrad at top energy. Assuming typical excitations, the orbit change expected due to a single COD failure is 1-3 mm (with 1-3 sigma confidence level). If the COD fails due to quench its field decays within 350 ms. If the power converter fails, the decay time is 60-80 s. It was noticed that the failure rate in LEP was not so different from this assumption, and that LHC is an environment with higher radiation level.
Now turning to optics errors, Ralph S. demonstrated that the orbit feedback works for up to 100% beta beating. Larger beta beating could not be studied in MAD as the optics became unstable.
Ralph
S. has built an orbit feedback controller test bed, which simulates the open
loop and orbit response and which allows offline debugging of the controller.
Most aspects of the feedback can thereby be tested and commissioned without
beam. Only a few points still require beam verification. These include
polarities, plane, and ring for BPMs and correctors.
For these tests, a circulating beam is needed and the coupling should be under
control. Afterwards, the feedback could immediately be operated, with a low
integral gain at 0.1-1 Hz. On the other hand reaching nominal feedback
performance requires a beta beating of less than 20%. This can be accomplished
by measuring the full orbit response matrix. Depending on whether the ORM is
measured automatically or manually between 4 and 8 h are needed in total.
Oliver wanted to know whether the same feedback is used for maintaining collisions. Ralph S. answered that another feedback will be used on top of the orbit feedback, based on the luminosity monitor, and that the coupling to this additional feedback is already foreseen in the architecture via the feedforward channel. Ralph A. recommended using the nearby BPM readings instead of the luminosity signal and setting tight limits on these readings, as during the commissioning the luminosity signal could be dominated by noise, a concern which was shared by Rhodri.
Ralph S. recalled that the orbit feedback does not attempt to correct energy errors, but the dispersive part of the horizontal orbit is subtracted prior to correction in order to minimize the cross-talk between orbit and energy feedback. Paul commented that this subtraction does not include the effect of spurious dispersion.
Jorg added that the feedback test in the SPS was very successful with a commissioning time of less than 1 shift. Ralph A. remarked that the feedback should be subjected to tests “under rough conditions”. Ralph S. described a procedure of exciting predefined oscillations at various frequencies and measuring the final amplitude, yielding a feedback transfer function, which could be used to compute the feedback response to arbitrary perturbations.
Rudiger highlighted that the orbit feedback is not part of the machine protection system. The latter is guaranteed by software interlocks. Ralph S. commented that combination of functions could indeed be dangerous.
The team of Verena and EICs has started following up the LHCCWG discussions on the web. The goal is developing well defined procedures, which will be used during commissioning. A draft web site was distributed prior to the meeting. The web site is structured in stages and phases. For the moment, the focus is on the form and less on the contents. Procedures for an example, namely the first turn, were already written as an illustration. Each commissioning phase is defined by entry and exit conditions for a number of subsystems, by procedures which allow reaching the exit, by problems and unsolved questions. Text in red is still to be discussed with the experts. Items listed under unsolved questions are to be reviewed in the LHCCWG.
Frank commented that the structure chosen would allow a matrix display, where one could easily extract orthogonal information, e.g., the commissioning sequence of a single subsystem. Verena replied that this had indeed been the intention following a suggestion from Brennan. Frank also remarked that some acronyms may not be easily understandable for all visitors to the web site. Ralph A. commented that the person or group responsible for each procedure should be indicated as well as an estimate of the time required. Roger agreed and added that for the procedural steps it should be spelt out exactly what needs to be done. Mike suggested splitting off all the sub-procedures. Many of these may be repeated in different phases. Oliver emphasized that maintainability should be a top concern, pointing to the LTC actions site as an example of the difficulties. Mike proposed introducing a higher-level navigation bar, going down to the lowest subdivisions. Each sub-procedure should have a unique label. All the systems involved need to be specified. The matrix display could be realized as a separate web page and via the navigation bar. Ralph A. recommended including in the milestones (here ‘exit conditions’) parameters like beta beating, intensities, etc. Stefano asked how to best follow up the open questions. It appeared some further thinking on the database organization could be beneficial for later maintenance and expansion. Roger encouraged all LHCCWG members to send feedback to Verena.
Wednesday May 31st, 14:30
CCC conference room 874/1-011
Provisional
agenda
Minutes of
previous meeting
Matters arising
Snapback and
ramp with single beam (Mike)
AOB
Reported by Frank