Noise simulation from data

Transcription

1 3x1x1 Analysis meeting 17/08/2018 Noise simulation from data (APC - Université Paris Diderot) scarpell@apc.in2p3.fr

2 Introduction 2 Noise in the 311 is not yet well understood Different detector conditions for the different runs Noise simulation is needed: Better understanding waveforms Validation for noise filters and hit finder Versatile noise simulation Adapt to all the noise configuration Easier and more realistic It doesn t tell much about the origin of the noise itself Outline for today: A quick noise study in the 311 Noise simulation strategy and performances A simple validation for the coherent noise filter

3 Source of noise 3 Principal noise contribution from SC cabling Noise RMS Noise FFT From the technical paper: High Voltage cables and temperature probes cables have a major impact Origin of the main frequencies is still under investigation 1 MHz burst not clear identified 200 khz freq. might be due to the instrumentation AC/DC converter 100 khz might be power supply

4 Noise time evolution 4 Run I Pulsing data HV tests and Run II Field scans July August September November 27/07-29/08 shutdown 07/09-05/11 shutdown Run /07 Caen issue 23/08 Elog 362 Noise increase not yet identified

5 Day night fluctuations 5 Day-night cycles doesn t seem to affect the noise Effect might be too small compared to the usual fluctuations Run 840 Run 842 Dusk Night Day

6 Noise frequency pattern evolution 6 Different noise Eras related to different detector conditions ( e.g. HV board replaced ). Different frequency patterns expected Noise simulation based on pedestals taken during different Eras Run Day Elog SC cables Henizinger CAEN LEMs Grid PMTs / Connected 0 kv 0 V Off Off Off / Connected 56 kv 0 V Off Off Off / Connected 0 kv 0V Off Off On

7 Channel Correlations reference pedestal for run I Raw data Correlation 32 ch correlation Correlation over larger multiple of 32

8 Frequencies map reference pedestal for run I Average FFT

9 Coherent noise simulation strategy 9 Correlation pattern Frequency map Model input Prepared outside larsoft Prepared using different techniques Support for different noise condition of the 311 Extract frequencies and amplitude for every channel Exclude dead and noisy channels Select only relevant frequencies (above a threshold) Each frequency get a random phase 0-2π Each frequency gets a random shift (-1, 1 rad ) Groups of corr. channels get the same shift Randomize amplitude Relevant frequencies are added as sinusoidal frequency on the interested channel Add an incoherent noise can be added LArSoft detsim stage Output: Simulated channel noise mask Slow: ~ 1.5 minute per event!!

10 Simulation performances 10 What is data? What is simulation?

11 Simulation performances 11 What is data? What is simulation? Data (run 840) Simulation (from July pedestal model)

12 Simulation performances 12 View 0 View 1 July pedestal (729-0) Noise simulation Average rms over 10 simulated noise events Noise level is comparable to the reference pedestal noise used for the simulation RMS value tends to be a little more stable than the reference Waveform example Non-optimal tune of Gain Tracks not perfectly at the same angle No slow component

13 Effect of the coherent noise filter 13 Coherent filter does an almost excellent job in removing most of the coherent noise frequencies Amplitude (Arb. units) 10 2 View 0 Before coh. noise removal After coh. noise removal Amplitude (Arb. units) 10 2 View 1 Before coh. noise removal After coh. noise removal f [MHz] f [MHz]

14 Effects of the coherent noise filter Before filter 14 After filter Wire View 0 Undershoot region Artifact of the CNR

15 Test for coherent noise simulation 15 Quick vanilla test of the coherent noise filter on simulation Before filter After filter Same artifacts as in data appears after CNR

16 Conclusion 16 Noise studies are ongoing A versatile noise simulation is in place Noise reference for different noise Eras has been identified Simulation works in LArSoft, but is a bit slow ( ~1.5 min per event ) Simulation looks quite realistic. No slow-baseline fluctuation modeled yet ( perhaps not needed ) Waiting for SC improvements for next MC production Coherent noise filter already exists since last year Effective reduction of the noise rms Artifacts mildly affects the signal It performs as expected also on the simulation

17 Thank you