Noise Mitigation and Simulation Jingbo Wang University of - - PowerPoint PPT Presentation
Noise Mitigation and Simulation Jingbo Wang University of California, Davis, Department of Physics CTF Meeting Feb 18 th , 2019 Topics Noise Mitigation in ProtoDUNE-SP Data Low pass filter Coherent Noise Removal Noise Simulation
Jingbo Wang University of California, Davis, Department of Physics
CTF Meeting Feb 18th, 2019
– Low pass filter – Coherent Noise Removal
– Noise simulation strategy – Noise modeling based on ProtoDUNE-SP data – Noise simulation service
Slide 2
Slide 3
Coherent noise High frequency pick-up noise
Slide 4
§ Do FFT analysis for waveforms with amplitude above/below threshold (20 ADC) § Coherent noise is noticeable at about 20-60 kHz. The source is known as the low- voltage regulator that power the cold electronics § High frequency pick-up noise appears at 310 kHz and 625 kHz § Very low-frequency noise comes from slow baseline fluctuations § Other APAs have similar coherent noise, but different high frequency noise
Slide 5
Slide 6
– PdspNoiseRemoval.h – PdspNoiseRemoval_tool.cc
– Makes groups by online DAQ channels (by chip or ASIC) – Makes groups by FEMB (40 U + 40 V + 48 Z) – Makes groups by offline channels (every N channel per group)
Slide 7
300 kHz cutoff frequency
Slide 8
200 kHz cutoff frequency
§ Some patterns can been seen over a long range of channels § Obvious groups: 16 channels per chip, 8 channels per ASIC, 128 channels per FEMB (40U + 40V+ 48Z)
– Grouping method is configurable – Every 16 channels per chip are grouped by default now
§ Mean or median waveform in each group is subtracted from channels § Channel status is checked: “bad” channels are not used § Signal (ADC>20) is protected (not included in mean/median calculation)
Slide 9
Slide 10
Run4696, APA#4, U plane
Slide 11
Raw waveforms After coherent noise removal
§ Significant correlation is clearly seen in a group of online channels. § After coherent noise removal, the correlation is much weaker.
Slide 12
Before After
§ Horizontal patterns are effectively mitigated by coherent noise removal
Slide 13
Before After
§ The peak at 30-40 kHz is gone after coherent noise removal § The induction planes feature a very low-frequency bin, which is related to the baseline oscillation.
Slide 14
Slide 15
§ Medium-frequency noise “spikes” can’t be removed by the low-pass filter § For some channels the noise “spikes” can be mitigated a little by the coherent noise removal, but for most of the time, this noise source is
§ A slow baseline fluctuation is seen in the data. The frequency is at the first several bins of the spectra. This is not removed at present § Signals (Amp>20 ADC) ROI is searched by a simple threshold
they don’t contribute to the coherent noise calculation. The performance of the noise removal slightly depends on ROI § If signals appear around the same tick, coherent noise removal doesn’t work
§ A new noise simulation service is added into dunetpc at
– dunetpc/dune/DetSim/Service/
§ The service is a combination of different noise generators. § The noise generation strategy follows the old generators:
1) Prepare the noise frequency spectrum for different sources 2) Do inverse FFT for each individual noise source to get noise waveforms 3) Add noise waveforms from all the sources to form a combined noise waveform 4) Generate a collection of noise waveforms at the service run time 5) Randomly pick up one noise waveform and add it to the simulated analog signal (AdcSignalVector) before digitization
§ As a cross-check, do FFT for the digitized waveforms (Rawdigit) to compare with the input FFT spectrum
Slide 16
§ Incoherent noise is fitted by the MicroBooNE noise model. § Coherent noise is obtained by subtracting the filtered FFT power spectra from the raw one. Then the magnitude is fitted by a exponential background and a series of Gaussian functions § Parameters are fed into the noise simulation service in dunetpc § Issue: very low frequency at the first few bins are not modeled well.
Slide 17
Thanks to Michael Mooney (CSU) MicroBooNE paper: https://arxiv.org/abs/1705.07341
Slide 18
White noise Incoherent noise MicroBooNE Noise Coherent Noise (Exponential + Gaussians)
Slide 19
ProtoDUNE-SP Data: Run4696 Larsoft Simulation
§ At present, the noise parameters are tuned according to the collection plane data. The same noise spectra is added to the induction planes. § The noise generator allows to generate incoherent high-frequency pick-up
present, such noise source is turned off § At present, the grouping of coherent channels is done for offline channel number (by default, 32 channels/group). It’s easy to add new functions to generate realistic groups. An offline-to-online channel map will be needed. § The coherent noise generator adopts the experimental frequency magnitude, but the phase information isn’t used. Needs to use the experimental phase information in the next update.
Slide 20
§ Prepared a noise mitigation tool to be added to DataPrep
– Low-pass filter (√) – Coherent noise removal (√) – High-pass and band-reject filters
§ Noise removal tested with the ProtoDUNE-SP data
– Old data: Run4696 – Recent data: work in progress
§ Prepared a new noise simulation service: – White noise, MicroBooNE noise model, Coherent noise model, Incoherent noise model – Parameters tuned with ProtoDUNE-SP data – New service keeps the old structure. Can be used for DUNE-FD – Need to use phase information for coherent noise generator
Slide 21