Meeting 03.05.2016
Participants: Dmitry, Doris, Adrian


  • Tel. resolution



  • enlarge the acceptance windows for patterns recognition and re-do the alignment
  • plot the GBL output errors for all planes w and w/o DUT as a scatterer
  • plot the RES_{GBL}^{x} and RES_{GBL}^{y} vs. x and vs. y
Analysis meeting 09.02.2016
Participants: Doris, Adrian, Dmitry


The distributions of the clustring-1 step and correlation plots (see Correlations paragraph) were discussed in a context of cuts:

  • Weird alternating behavior of the strip signals distr. vs. alibava ch. (see fig. 1, left): the av. signal of a strips group is negative while the following group of strips has positive av. signal (signals dist. was plotted taking to account negative signal polarity from the MCz200P sample). This behavior has to be investigated in details to verify the data. It would make sense to plot the raw charge dist. vs. ch. and compare it to the distr. after the common mode correction (the cross-talk wasn't taken to account yet, its effect is on a few percent lvl., so in the second order correction) noise on fig. 1.

Fig. 1. Signal and noise distr. vs. alibava ch. for 50000 ev. of the run000013.dat, SeedSNRCut = 2.0

(MCz200P(irr. 1.3 \cdot 10^{16}cm^{-2})@-1000V, 0deg. incl., -30deg.C, before annealing).

  • Cluster Charge vs. TDC distr. (fig. 2) demonstrates the "Gaussian bell"-like shape as it's expected to be with accordance to the beetle chip signal's shape.

Fig. 2. CluCharge vs. TDC for 50000 ev. of the run000013.dat, SeedSNRCut = 2.0.

!NB This distr. shouldn't start from 0 along Y axis with a SeedSNRCut being set because the cluster charge averaged upon its strips shouldn't include any noise component since it's being cut by the NeighborSNRCut criterion (positive and negative charge of the same abs. values give zero after summing/averaging). So the cluster charge calculation should be corrected to cut out the noise charge below the cut lvl. (see example on fig. 3).

Fig. 3. CluCharge vs. TDC for 11000 ev. of the run000183.dat (June15 run for FTh200P non-irr. module@-250V, 0deg. incl., -25deg. C), SeedSNRCut = 5.0,

Thomas' code with the old EUtelescope.

  • Strips with higher occupancy were observed on the correlation plots (see fig. 4). This has to be clarified, not well understood yet; it could that these patterns somehow related to the influence from the noise neighboring strips, but has to be verified by telescope tracking.

Fig. 4. Cluster correlation plots along Y axis, CMSPixRef vs. DUT; 50000 ev. of the run000013.dat, SeedSNRCut = 2.0

(MCz200P(irr. 1.3 \cdot 10^{16}cm^{-2})@-1000V, 0deg. incl., -30deg.C, before annealing).

  • SNR for this irr. sample is ~5.5-6.8 (see fig. 5.44 in the Thomas' thesis, the values on the plot were obtained after the tracking procedure), but no peaks at this lvl. were observed in the ClusterSNR distr. (fig. 5). This doesn't necessarily mean that the data we took are useless, but does require the following checks:
        1. whatever the hist. on fig. 5 was filled right/cluster SNR calculation was correctly done
        2. if it's so, tracking procedure for the telescope has to be performed to find which exactly strips are being fired at a particular event and find the real signal clusters, because it seems like the signal is too low and covered by the noise background. At least the eta distr. gives a hope for signal presence, see fig. 6.

Fig. 5. Cluster SNR dist. for 50000 ev. of the run000013.dat, SeedSNRCut = 2.0

(MCz200P(irr. 1.3 \cdot 10^{16}cm^{-2})@-1000V, 0deg. incl., -30deg.C, before annealing).

  • Right plot on the fig. 6 shows better peaks separation corresponding to the absence of charge sharing when a sensor oriented perpendicularly to the beam, while the left plot shows higher peaks overlap related to the noise signals which degrade position sensitivity of the detector.

Fig. 6. \eta dist. for 50000 ev. of the run000013.dat; left - SeedSNRCut = 2.0, right - SeedSNRCut = 5.0.

!NB Check the \eta calculation and exclude clusters which size is greater than 2.


  • Check the charge sign while clustering and correct the cluster charge distr. to omit the signal lower than the cut
  • Clean up the \eta distr. to take the only clusters which size is 2 
  • Compare the raw charge distr. vs. ch. to the charge after the noise subtraction distr. vs. ch.

  • Hitmaker: what does it do, which data collections uses etc. write the details and accomplish this step -> p. 3.4.5 of the Thomas' thesis.
  • Focus on the telescope alignment w/o DUT in order to find the tracks and select the DUT region where strips are being fired and not just accumulate noise BG signal


  • Next TB analysis meeting will take place on Mon., 22.02.2016 at 9.00. Place to be defined.
  • Dmitry's presentation at PhII Upgr. meeting should be preliminary scheduled on 04.03.2016.


Files looked at :



root distr. for run000013.dat

(MCz200P(irr. 1.3 \cdot 10^{16}cm^{-2})@-1000V, 0deg. incl., -30deg.C, before annealing)


run000013-merger_SNR2.0.root, 000013-alibava-clustering-1_SNR2.0.root
5.0/3.75run000013-merger_SNR5.0.root, 000013-alibava-clustering-1_SNR5.0.root
 0.0/0.0 000013-alibava-clustering-1_SNR0.0.root




Analysis meeting 26.01.2016
Participants : Adrian, Dmitry

Last week's TODO list discussion

  • The plots aka telescope DQM showing the av. numb. of clusters, cluster X and Y size per telescope+CMSPixRef, the av. numb. of clusters per telescope arm vs. time/run number (functions f_{upstream}(t) and f_{downstream}(t) ) as well as the difference and asymmetry of these functions,  f_{downstream}(t) - f_{upstream}(t) and \frac{f_{downstream}(t) - f_{upstream}(t)}{f_{downstream}(t) + f_{upstream}(t)} are presented (w/o fiducial cut) on the page Telescope raw data analysis.
  • Merging step was clarified, two codes, git ver. and Thomas' ver., were compared (see the merge step):
    1. Both codes have the same purpose of merging the pulse and sparse clusters' collections of the telescope and alibava data to a one *.slcio file.
    2. Git ver. code is easier to read through; it doesn't contain functions for drawing histograms in the contrary to the Thomas' code which allows to draw correlation plots (just in case when CMSPixRef in off though).
    3. It's unclear, how the alibava input *.slcio file in being read: the variable containing the file name is being defined, but never used in the code or being sent to another processor -> Dmitry will meet with Thomas next week to discuss the code.
  • The git ver. merging code was used for merging the telescope+DUT+CMSPixRef data with the following cuts which were applied on the clustering-1 step (git ver. code was used for that step):
  1. TDCCut = 0-100 ns (TimeCutMin = 0; TimeCutMax = 100)
  2. SeedSNRCut = 2
  3. NeighbourSNRCut = 1.5.

The plot showing the fraction of events which contain clusters per each telescope+DUT+CMSPixRef vs. bias HV/anneling time and inclination angle as the result of the merging step is presented on the page Alibava data, titled as "After the merger step". The av. fraction lays within 39-45%; that is in a quite agreement with the geometrical acceptance of the CMSPixRef which is ~45-46% (with the physical  limitation of 50%). Lowering of the fraction value could be related to a slight mis-tunning between the DUT and CMSPixRef timings.


Quick glance though the clustering-1 code:

  • Thomas' code ( is less straightforward for understanding that the git ver. code (
  • Thomas' code, as its explained in his thesis (p. 5.3.4 ,, has the cross-talk noise correction, while git ver. one doesn't
  • SNR cuts have the same purpose in both versions.


  • Clustering step. Compare the code versions and modify existing one, e.g. for being able to make the cross-talk noise charge correction
  • Play with the cuts :
  • Define the optimal TDC cut/time selection window
  • Define the optimal SeedSNRCut and NeighbourSNRCut parameters - N vs. SeedSNRCut, N vs. NeighbourSNRCut
  • Compare fractions of the events where the clusters were found on each telescope+DUT+CMSPixRef plane calculated for the Oct15 runs  with the ones calculated for the June15 runs - irr. vs. non-irr. sample comparison:
  • to fix the TDC cut to 0-100 ns
  • to vary the SeedSNRCut and NeighbourSNRCut parameters

and see the dependence "Fraction vs. SeedSNRCut vs. NeighbourSNRCut" in the following cases:

  1. Oct15 run before the sample was annealed vs. Jun15 run; inclination angle is being fixed to 0 deg., temperature is being fixed to the lowest one reached (-30/-25 deg.C for the Oct15/June15 runs)
  2. Oct15 run after the sample was annealed vs. Jun15 run; inclination angle is being fixed to 0 deg., temperature is being fixed to the lowest one reached (-30/-25 deg.C for the Oct15/June15 runs)
  • Think about the questions which can be asked during the Telescope Workshop at Orsay -> send them to Adrian

Analysis meeting 19.01.2016
Participants : Doris, Dmitry, Adrian, Hendrik 



  • DAQ for the telescope data taking to account the time when the data runs were performed. why do av. cluster numb. change (e.g. wrong mimosa26 threshold or the beam intensity/energy which corresponds to the noise). compare the av. cluster numb. for a certain plane, for all planes. Check the same for cluster size x and y, it should be checked as a function of time. Plot the difference between 0-2 and 3-5. (remark: exact time when the runs occurred is not known. would it be possible to check the Edaq logs the see when it was launched and extract the run time info?)
  • Fiducial (volume)/acceptance cuts: on the hitmat of the plane2, cut out the clusters which are laying on the plane in the region out of "the DUT shadow"/DUT projection onto the plane2
  • extra-plots for CMSPixRef to see the difference

Idea for the next TB: we should take a (few) run(s) w/o DUT physically mounted to the telescope to see its ifluence on scattering/photon creating etc.

Analysis meeting 12.01.2016
Participants: Dmitry, Doris, Hendrik

Telescope raw (run000224.raw)

The numb. of clusters per plane were calculated using the processors for alibava data analysis w/o including the DUT and CMSPixRef and the datura-noDUT processors/steering templates (10000 events were taken to account):

IDTotalPer event






The numb. of clusters per plane using the processors for alibava data analysis with CMSPixRef (50000 events were taken to account):

IDNumb. of clustersNumb. of clusters per event

These calculations show that the CMSPixRef acceptance is eq. to ~49% which agrees with TB's value. That means that the telescope data are OK, and low merging efficiency (was reported during the prev. meeting) can be related to the fact that the amount of clusters per Mimosa plane per event is two orders of magnitude higher than the numb. of clusters on CMSPixRef per event. This difference in the cluster numbers can be led by noise signal at Mimosa sensors.

Comparison between the Thomas and GitHub versions' steering-templates:

similar analysis steps

    - convert-ped

    - pedestal

    - commonmode

    - pedestal2

    - converter

    - reco

    - clustering-2

The template clustering-1 differs, so the one from the git-hub is being used since it was working properly. Another clustering-1 template was prepared based on the Thomas template, it's called clustering-1-thomas-tmp.xml.
The template telescope-converter differs as well: the Thomas ver. has HotPixelKiller processor which executes the; the git-hub ver. has the HotPixelMasker processor executing the file.
The template telescope-clustering differs auch: the git-hub ver. has the ClusteringCMS processor.

Processors' *.cc files

The github's ver. with the Eda's commonmode noise estimation approach was "frankensteined" with the Thomas' ver. The following files were taken from the Thomas' ver.:
- -> for the convert-ped step
-,, -> for the commonmode step.
The step pedestalhist wasn't changed much with respect to the Thomas' ver., the only change is the method histogras are being put to the root tree. doesn't differ between the versions.


  • plot histos with numb of clusters and cluster size per telescope plane as functions of time (event) for several runs, and check the run to run trend Unknown User (kislerdm)
  • look to the mergin step and document how the cuts (e.g. TDC cut) are being set there Unknown User (kislerdm) 
TB Meeting 16.12.2015
Participants: Dmitry, Doris, Alex, Adrian

Planning for TB'16

  • TB21 is presumably going to be booked for us (Doris Eckstein) for the weeks 23 and 24 (June, 6-19th), 42 and 43 (Oct., 17-30th);
  • Ppls for this TB: Dmitry, Doris, Adrian, Paul (80% sure), Matteo (50% sure), Anastasia /new PhD student of Doris and Hendrik/ (50% sure);
  • Sensors to be measured: Epi100P(irr. 1.3\cdot10^{16} cm^{-2}) and ... ;
  • Two weeks in prior to TB have to be taken for lab tests;
  • Meeting (Dmitry and Adrian) with Alex on 20.12 at 10.00 by the Alex' office;

  • Dmitry is at CERN for the Trk DOC shifts during the weeks 13 and 16 (27.03 - 04.04.2016 and 17-25.04.2016).

Thoughts about sensors for future tests

  • The sensors' size for future tests will be 1\times1\ cm^2, PCB size is 5\times5\ cm^2. Phase I guys can be asked for these PCBs because they use exactly them. NB! PCB design and manufacturing takes up to 3 months, so one has to think quite in advance to the experiments/TB/etc.
  • Box for future TB tests should be modified, we can have a look to the box which is in use for the Phase I tests and get some ideas for us;
  • We shall probably transport the TB equipment (e.g. supplies, chiller etc.) to the DESY e-lab to avoid shipment issues each time we want to have beam time.


  1. High number of clusters on the telescope was observed when no clusters were detected on DUT and CMSPixRef. This can be related to the high DAQ rate which was observed during the beam time. Such events should be excluded or maybe used for telescope alignment.
  2. Clustering efficiency, i.e. the ratio between the number of events when clusters were detected on telescope as well as on the DUT and CMSPIxRef. with respect to all events from the considered data sample, is ~2%/199 out of 10000 events/ (run000012.dat, run000013.ped, run000224.raw, sample@-1000V, 0 deg. inclination, no annealing, -30 deg. C) - too low! E.g. the efficiency for the June's data was up to 30%. This can be related to used TDC cuts (from 25 to 45 ns - these cuts were set as default in the steering-tamplate, so wasn't any reason to use them).