Gaussian Deconvolution Filter Michael Mooney Brookhaven National - - PowerPoint PPT Presentation

Gaussian Deconvolution Filter

Michael Mooney

Brookhaven National Laboratory ProtoDUNE-SP DRA Meeting – June 8th, 2017

♦ In ProtoDUNE-SP reconstruction, currently a “1D” deconvolution is utilized, along with a Wiener filter

• 1D deconvolution: not taking into account signal induction on

wires neighboring the wire closest to ionization electrons

• Wiener filter: deconvolution filter that gives maximal peak-to-peak

separation (“sharpness” of image)

♦ Most ideal to use 2D deconvolution (FFT over both time domain and wire domain, then remove 2D response)

• BNL group (Xin et al.) has agreed to bring this technique to

ProtoDUNE-SP

• First releasing/tuning for MicroBooNE (~August) so this will come

later for ProtoDUNE-SP (by end of the year?)

♦ Until 2D deconvolution ready, since we use GaussHitFinder, we should move to a Gaussian filter – focus of this talk

2

Introduction Introduction

♦ MicroBooNE has released public note documenting signal processing techniques useful for LArTPC experiments

• See public note here: MICROBOONE-NOTE-1017-PUB

♦ This note describes 2D deconvolution technique

• Technique improved since public note – paper forthcoming
• Nature of detector response different than current assumption that
• nly closest wire matters (see below figure)
• Worst for MicroBooNE (3 mm spacing), still important for PD-SP

3

MicroBooNE SP Public Note MicroBooNE SP Public Note

♦ MicroBooNE has released public note documenting signal processing techniques useful for LArTPC experiments

• See public note here: MICROBOONE-NOTE-1017-PUB

♦ This note describes 2D deconvolution technique

• Technique improved since public note – paper forthcoming
• Nature of detector response different than current assumption that
• nly closest wire matters (see below figure)
• Worst for MicroBooNE (3 mm spacing), still important for PD-SP

4

MicroBooNE SP Public Note MicroBooNE SP Public Note

See Backup Slides For 2D Deconvolution Examples at MicroBooNE

♦ Also detailed in the note is the importance of a deconvolution filter – prevents noise blow-up when dividing out response ♦ Wiener filter gives optimal peak-to-peak separation, but we are fitting to Gaussians (GaussHitFinder) → problems!

5

Signal Processing Concepts Signal Processing Concepts

Deconvolution Wiener Filter

6

MicroBooNE U Plane MicroBooNE U Plane

7

MicroBooNE V Plane MicroBooNE V Plane

8

MicroBooNE Y Plane MicroBooNE Y Plane

♦ If fitting post-deconvolution signal with Gaussians (GaussHitFinder), would overestimate charge with Wiener filter (peaks of integral plots) – for MicroBooNE (273 V/cm):

• U Plane: overestimate charge by ~14%
• V Plane: overestimate charge by ~19%
• Y Plane: overestimate charge by ~18%

♦ Advantages of Gaussian filter:

• Minimal bias charge extraction when using Gaussian fits
• Better goodness of fit
• Uniformity of dQ/dx vs. angle of track w.r.t. anode plane
• No noise hits from “side bumps” associated with Wiener signal

♦ Simple to switch in code (.fcl file parameters)

• First need to study/optimize for ProtoDUNE-SP

9

Discussion Discussion

10

BACKUP SLIDES

11

2D Deconv. Example 1 2D Deconv. Example 1

12

2D Deconv. Example 2 2D Deconv. Example 2