15
IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

Embed Size (px)

Citation preview

Page 1: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012Slide 0

EMMA: Update on BPM modelling and mapping

March 1st 2012

Ian Kirkman

Page 2: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012

Brief reprise of need for mapping of EMMA BPMs

Slide 1

• Electrode response does not vary linearly with distance to beam• Response is a function of the BPM geometry and materials• Cannot use 4-button responses to simply deduce beam position – errors of up to > 5mm in recovered beam position can arise• Therefore the particular BPM response needs first to be measured or modelled

Page 3: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012

Measuring and Modelling the EMMA BPMs (2010 onwards)

Slide 2

Bench tests

Theory

Electrostatic(2-D)

CST(3-D)

Four approaches essentially used to confirm the reliability of the CST method. CST data then used for all further modelling.

Page 4: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

First look was in early 2010:A 7th order polynomial based mapping function used to cover entire(Xunc, Yunc) space

More points added for cylindrical 2011/12 : Essentially a 4mm grid adopted over area of approx. linearity

But points get denser points as beam moves away towards the beampipe edges

Too few points modelled for rectangularBPM assembly – needs more

Use of new “local fit” method appears to necessitate a finer mesh of modelled points – in progress.

IK March 1st 2012

Status of CST modelling

Slide 3

Page 5: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012

Some details of CST modelling

Slide 4

• Particle Studio used (identical results obtained with e.g. EM Studio)

• DC system assumed (confirmed okay through bench tests)

• Full double BPM beampipe assembly modelled “as one” in 3-D

• “Cylindrical” and “Tapered Cylindrical” assemblies give subtly different results

• CST does not use paging memory – so limited by RAM of machine (few Gbytes)

• Approximately 5 million hexagonal meshcells modelled, with automatic mesh generation

• If uniform mesh, mesh cell ~ 400 um on a side, but actually much smaller where model requires it for specified 1 in 10^6 accuracy

• Each point takes approximately 1 hour to run

• Cannot set up batch runs – limited user licence server – frowned upon

Page 6: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012

Use of modelled data for subsequent mapping

Slide 5

Xunc

Yunc

Xunc

Yunc

[Alex. Kalinin]

The values of Xunc and Yunc are then used in the mapping to recover the Xreal, Yreal values

Page 7: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012

CST data for Xreal vs (Xunc, Yunc) – Cylindrical BPM

Slide 6

Page 8: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012

CST data for Yreal vs (Xunc, Yunc) – Cylindrical BPM

Slide 7

Page 9: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

We assume the bunch charge is proportional to the sum of the 4 electrode responses(see, for example, “Charge, position and resolution computing in EMMA BPMs”, Alex. Kalinin, EMMA meeting, 14-09-11)

Suppose a bunch at the beam centre gives ∑Vi = 1

The same bunch situated close to an electrode would give ∑Vi >> 1

Therefore, we cannot simply use ∑Vi as a measure of the charge (or variation of charge) in the circulating bunch

Approach adopted is to simulate a unity charge at a variety of x and y positions, then to provide a normalisation value, Qf, which must be divided into ∑Vi at any beam location to derive the bunch charge which would have been measured if the beam had been at the centre of the beampipe

See following figures

IK March 1st 2012

Charge Normalisation Factor, Qf

Slide 8

Page 10: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012

CST data for Qf vs (Xunc, Yunc) – Cylindrical BPM

Slide 9

Page 11: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

Procedure up to present: Fit a single 2-dimensional 7th order polynomial to each of the surfaces in slides 6,7 and 9 Calculate Xunc, Yunc from real-data button responses (‘scope or EPICS), correcting for “tails” Interpolate to the appropriate surface, giving Xreal, Yreal or Qf Advantage: gives a smooth mapping function Disadvantage: polynomial doesn’t fit well near to electrodes (and higher order produces

extraneous peaks and troughs)

New “Local-Fit” procedure: Calculate Xunc, Yunc from real-data, as above Identify local region of surface around Xunc, Yunc (e.g. nearest 25 modelled points) Perform a 2-D 2nd order polynomial fit to this restricted dataset Either: use fit parameters to interpolate Xreal, Yreal values, or

pre-calculate look-up tables (e.g. every 50 um) giving Xreal, Yreal, Qf for any combination of Xunc, Yunc

Advantage: MUST produce a better fit to the modelled data than fitting to whole (-1,+1) surfaceDisadvantage: more susceptible to “steps” in modelled data – need denser modelled points

IK March 1st 2012

Mapping procedures

Slide 10

Page 12: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012

Mapped data for Xreal, Yreal and Qf – Cylindrical BPM

Slide 11

Xreal Yreal QfCompare mapped (fitted) surfaces using “local-fit” with original modelled data in slides 6,7 and 9:

RMS deviation of fit from X and Y data ~< 100 umRMS deviation of fit from Qf data ~ 0.25

The RMS uncertainties in X and Y (and particularly Qf) will be considerably improved by modelling more points using CST – this is in progress

Page 13: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

Cylindrical assemblies:Electrode responses L and R subject to same amplification “gain”; responses U and Dpotentially subject to a different gain. If gain for L,R is 1, and gain for U,D is “g”, can write:

(So different channel gains have no effect on recovered positions)

Rectangular assemblies:

[after Alex. Kalinin]

Simple differentiation w.r.t. g gives errors in Xunc and Yunc in the sub- 10 micron range for g < ~0.1 and typical values of A, B, C and D

So for expected maximum channel gain differences, errors arising are << e.g. thermal noise, and this is borne out by mapping analyses

IK March 1st 2012

A note about channel gains

Slide 12

Page 14: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

Difficult to gauge quantitatively, but contributions from:

Modelling: - accuracy of CST representation of assemblies - accuracy of interpolation (using mesh of finite size) - mesh size < 400 um

Mapping: - accuracy of global polynomial or local-fit approach to interpolate between modelled points

typically < 100um for X and Y, and ~ 0.25 for Qf (see slide 11)

Need to: - Look again at CST auto mesh generation more carefully - Re-run many more points under CST and repeat mappings

IK March 1st 2012

Uncertainties arising from the mapping approach

Slide 13

Page 15: IK March 1st 2012 Slide 0 EMMA: Update on BPM modelling and mapping March 1 st 2012 Ian Kirkman

IK March 1st 2012

EPICS in use

Slide 14

• Only the essential “raw” data is stored in the EPICS database• All processing and analysis of this raw data, including tail corrections, electrode responses, mappings, and other corrections (e.g. offsets) to be done on command

by a LINUX based server box•Peripheral software, “real-time” or off-line, can interrogate the LINUX box with a time

stamp to obtain real X, Y and Qf values for any cell• If such processing were done on the cards themselves:

- slower response- not “future-proof” to changes in e.g. offsets or mapping procedure- any errors in processing would be irrecoverable