R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Much Noise About Nothing? Robert Zitoun Stony Brook and LAPP CTF Meeting February 10, 2003
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Channel noise ADC noise depends on the cell size and/or the preamp: Larger cells -> larger noise Coarse hadronic Amplified by cal_weights (total to visible energy ratio) EM/HAD ~10 MeV/ADC CH ~ 30 MeV/ADC ×1.6 for non linearity Noise/ADC Noise/GeV
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February st line = raw noise/ADC 2 nd = cal weights 3 rd = noise/GeV >100 MeV Channel noise ~40 70 ~ very high noise
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Occupancy Zero 2.5 gaussian noise 1% a lot of good channels BUT bad channelsugly channels Remaining noise above 2.5 is now at least EM/HAD ~ 300 MeV CH ~ 750 MeV for 450 runs 1% ← 15% ←100%
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Bad Channels Should be caught as early as possible Shifters should kill high occupancy channels and they should be replaced quickly 30 rms 10- occupancy 100% online pedestal 40 ADC online 2.7—
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Bad Channels – Effect on MET Killing bad channels does good for MET before ( ) and after ( ) removal (METx or y) before ( ) and after ( ) removal METxMETy
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Ugly channels High occupancy ~10-15% Several hypotheses studied Pedestal shift Non gaussian noise L1 SCA dispersion + updown difference Special calibration run
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Pedestal Shift (1) online pedestal 6ADC 1ADC = 3.1 = –2.1 lower roundingupper rounding
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Pedestal Shift - Macroscopic Effect –sum of rounded pedestals in a card measured SET in card
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Pedestal shift does not explain ugly channels with high occupancy 10 rms 10- occupancy 10% online pedestal 2 ADC online 4—4— Pedestal shift (flop!)
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Gaussian Noise Non gaussian noise ? Compare histo rms to gaussian fit channel 52 (-15, 33, 3) rms = 2.17 = 2.14 -rms ↑rms binning effect? ??
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Non gaussian noise Fix these channels, but what’s the problem?
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February L1 SCA dispersion Noise comes mainly from calorimeter/preamp Also from cell to cell dispersion in L1 SCA (much less in L2 SCA) ADC counts cell # in L1 SCA
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February L1 SCA up-down difference More hardware to fix!!! up down
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Offline Zero Suppression Data taken with 1.5 zero suppression from individual calibration runs Reco suppresses at 2.5 from a special run taken on ~September 20 ~ 50 k events to average out any L1 SCA problems Reco suppression with histo rms MC generation with gaussian fit → MC hot cells
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Offline Zero Suppression is sometimes wrong! offline suppression 2.5 × 2 = 5 ADC < online 1.5 × 4 ADC = 6 ADC Same effect seen in most channels with high (~10%) occupancy good channelbad channel ↑special run pedestal noise normal noise level→ (symmetrized noise)
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Suppression with 2.5 × online Implement 2.5 cut on thumbnails and look at MET (bad cells not removed) Certainly smoother distributions with lower rms before after
R. Zitoun, Stony Brook and LAPP CTF Meeting10 February Last Transparency, Mr Chairperson Present calorimeter OK for x-section measurements But lot of effort still needed to fix the calorimeter (change boards and/or daughter cards) understand its data (calibration strategy) improve its resolution (not addressed here!) Prepare a tool to clean the data (2.5 cut, kill bad cells, ped shifts,...) Would be worth trying it on “noise” jets Please do not drop CAL block from thumbnails before cal is OK Let’s try a 2.5 online cut