Background Suppression with Trigger Neuro Team Suppression - - PowerPoint PPT Presentation

▶
background suppression with
SMART_READER_LITE
LIVE PREVIEW

Background Suppression with Trigger Neuro Team Suppression - - PowerPoint PPT Presentation

Dec 16, 2023 99 likes •505 views

Background Suppression with Trigger Neuro Team Suppression Simulation Background Algorithm NeuroTrigger Belle II the Belle II Neural Network Introduction Outline Mar 19, 2018 Max-Planck-Institut fr Physik Sebastian Skambraks Trigger

slide-1
SLIDE 1

Background Suppression with the Belle II Neural Network Trigger

Sebastian Skambraks

Max-Planck-Institut für Physik

Mar 19, 2018

Outline

Introduction Belle II Trigger NeuroTrigger Algorithm Background Simulation Suppression

Neuro Team

  • S. Bähr, C. Kiesling, S. Pohl, S. Skambraks
slide-2
SLIDE 2

Introduction - Belle II at SuperKEKB

located in Tsukuba, Japan at KEK

高エネルギー加速器研究機構

K¯

  • Enerug¯

ı Kasokuki kenky¯ u kikou High Energy Accelerator Research Organization

e+ 4 GeV e− 7 GeV

  • asymmetric e+ e− collider
  • Υ(4S) resonance
  • B0 B0/ B+ B−
  • L = 8 × 1035 cm−2 s−1

(40× KEKB)

  • average pT: 500 MeV
  • average track multiplicity: 11

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 2/ 14

slide-3
SLIDE 3

Introduction - The Belle II Detector

e− e+

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 3/ 14

slide-4
SLIDE 4

Introduction - The Belle II Detector

e− e+ Central Drift Chamber 56 layers Input for L1 Track Trigger

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 3/ 14

slide-5
SLIDE 5

Introduction - Belle II Background

Beam Background Tracks

p h y s i c s background e− e+ z

  • tracks generated at the beam-line & -wall

with vertices z = 0 cm

  • increase with luminosity
  • main processes:
  • Touschek efgect
  • radiative Bhabha back scatters
  • beam gas

NeuroTrigger Goals

  • reject tracks from z = 0 cm
  • single track z-vertex resolution < 2 cm
  • latency < 1 µs

z (cm)

  • 40
  • 30
  • 20
  • 10

10 20 30 40 # of events / 5 mm 200 400 600 800 1000 Z distribution

Belle ⇒ need z vertex reconstruction at 1st trigger level

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 4/ 14

slide-6
SLIDE 6

Introduction - Belle II First Level Trigger

CDC ECL KLM PID GDL 30 kHz tracking

5 µs

Requirements

  • 30 kHz trigger rate
  • 5 µs latency

⇒ deadtime-free pipelined operation

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 5/ 14

slide-7
SLIDE 7

Introduction - Belle II First Level Trigger

CDC ECL KLM PID GDL 30 kHz tracking

5 µs

Requirements

  • 30 kHz trigger rate
  • 5 µs latency

⇒ deadtime-free pipelined operation

CDC Trigger Pipeline

CDC

  • 1. TSF
  • 2. Finder
  • 3. Tracker

GDL Hough Transformation (2D Tracks) Neural Network (3D Tracks) Track Segment Finder

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 5/ 14

slide-8
SLIDE 8

Introduction - CDC Trigger

5 axial super layers 4 stereo super layers

≈16 cm ≈ 1.2 m z ≈2.4 m

axial layer stereo layer

  • 56 layers combined to 9 super layers (SL)
  • 2336 track segments (TS) in 9 SL

SL angle (mrad) 2 45.4 – 45.8 4

  • 55.3 – -64.3

6 63.1 – 70.0 8

  • 68.5 – -74.0

Stereo SL confjguration

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 6/ 14

slide-9
SLIDE 9

Introduction - CDC Trigger

5 axial super layers 4 stereo super layers

≈16 cm ≈ 1.2 m z ≈2.4 m

axial layer stereo layer

  • 56 layers combined to 9 super layers (SL)
  • 2336 track segments (TS) in 9 SL

SL angle (mrad) 2 45.4 – 45.8 4

  • 55.3 – -64.3

6 63.1 – 70.0 8

  • 68.5 – -74.0

Stereo SL confjguration

Track Segment

≈ 15 mm

NeuroTrigger Input

  • position, drift time and left/right

information of TS priority wires

  • 2D track estimates (pT, ϕ)

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 6/ 14

slide-10
SLIDE 10

Introduction - CDC Trigger

  • axial layers
  • stereo layers
  • Υ(4S) event
  • background noise
  • track segments (TS)
  • 10

10 x [mm] 150 300 t [ns] xt – relation (nonlinear)

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 7/ 14

slide-11
SLIDE 11

Introduction - CDC Trigger

  • axial layers
  • stereo layers
  • Υ(4S) Event
  • background noise
  • rack segments (TS)
  • 10

10 x [mm] 150 300 t [ns] xt – relation (nonlinear)

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 7/ 14

slide-12
SLIDE 12

Introduction - CDC Trigger

  • axial layers
  • stereo layers
  • Υ(4S) Event
  • background noise
  • track segments (TS)
  • 10

10 x [mm] 150 300 t [ns] xt – relation (nonlinear)

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 7/ 14

slide-13
SLIDE 13

Introduction - CDC Trigger

  • axial layers
  • stereo layers
  • Υ(4S) Event
  • background noise
  • track segments (TS)
  • 10

10 x [mm] 150 300 t [ns] xt – relation (nonlinear)

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 7/ 14

slide-14
SLIDE 14

NeuroTrigger - Multi Layer Perceptron

Properties

  • robust function approximator
  • massively parallel processing
  • short deterministic runtime
  • neuron: y = tanh(wixi + w0)
  • network: zk = f (wkjf (wjixi))

Training

  • minimize

i

  • z

True

i

− z

Net

i

2

  • RPROP (backpropagation)

input one TS Hit per SL per track (position ϕrel, α and time t)

  • utput z, θ estimate

z θ . . . . . . wkj wji tsl ϕrel

sl

αsl tsl ϕrel

sl

αsl tsl ϕrel

sl

αsl tsl ϕrel

sl

αsl tsl ϕrel

sl

αsl

input layer hidden layer

  • utput

layer

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 8/ 14

slide-15
SLIDE 15

NeuroTrigger - Input Representation

axial stereo U axial stereo V axial stereo U axial stereo V axial drift time t ϕrel arc length s

ϕrel : TS position relative to 2D track α : 2D arc length to TS r2D

  • use track estimates provided by 2D fjnder
  • 3 inputs per SL, values: (t, ϕrel, α)
  • dedicated networks for missing hits

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 9/ 14

slide-16
SLIDE 16

Background Simulation

background process 2D trigger rate TwoPhoton e+e− →

  • e+e−γγ

e+e−e+e− 6 kHz Bhabha S 20 kHz Bhabha M e+e− → e+e−γ 52 kHz Bhabha L 26 kHz Touschek intra bunch scatt. 2 kHz Coulomb e±N → e±N 15 kHz Brems e±N → e±Nγ 1 kHz

Luminosity Machine

Triggered Particles

e + e

+

p

  • ther

particles 20 40 60 kHz Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

Track Multiplicity

1 2 3 tracks 20 40 60 80 100 120 kHz

θe+[◦] θe−[◦] 0.5 1 10 0.51 10

BhabhaS BhabhaM BhabhaL

3 Bhabha cases

(In the dominating t-channel, the Bhabha cross section strongly depends

  • n the scattering angle)

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 10/ 14

slide-17
SLIDE 17

Background - Material Scattering

Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

Initial Bkg Particles before Scattering

150 100 50 50 100 150 z / cm 20 40 60 kHz rate: 121.5 kHz

  • primary generated

bkg particles

  • only events with a

2D trigger

  • luminosity bkg
  • nly from the IP
  • machine bkg

from the beam pipe

150 100 50 50 100 150 z / cm 5 10 15 20 25 30 r / cm

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 11/ 14

slide-18
SLIDE 18

Background - Material Scattering

Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

Tracks seen in the Trigger

150 100 50 50 100 150 z / cm 5 10 15 20 25 kHz rate: 121.5 kHz

  • particles after

detector simulation

  • bkg particles matched to

2D trigger tracks ≈ 80 kHz reducible (z = 0) ≈ 40 kHz irreducible (z = 0)

150 100 50 50 100 150 z / cm 5 10 15 20 25 30 r / cm

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 11/ 14

slide-19
SLIDE 19

Background - Material Scattering

Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

Reducible Background Tracks

150 100 50 50 100 150 z / cm 2 4 6 8 kHz rate: 81.4 kHz

  • particles after

detector simulation

  • bkg particles matched to

2D trigger tracks ≈ 80 kHz reducible (z = 0) ≈ 40 kHz irreducible (z = 0)

150 100 50 50 100 150 z / cm 5 10 15 20 25 30 r / cm

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 11/ 14

slide-20
SLIDE 20

Background - Reconstruction

Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

Neural Network Track Estimates

150 100 50 50 100 150 z / cm 0.0 2.5 5.0 7.5 10.0 12.5 kHz rate: 76.2 kHz

  • 3D reconstructed bkg

with the neural network

  • neuro z range limited

to [−50, 50] cm

100 50 50 100 zMC / cm 100 50 50 100 zNN / cm

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 12/ 14

slide-21
SLIDE 21

Background - Suppression

Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

Z Cut (Tracks not from IP)

10 20 30 40 50 z cut / cm 20 40 60 80 100 120 kHz

  • only tracks with

|zMC| ≥ 1 cm

  • cumulative bkg rate

after a cut on the neural network z

  • zcut is varied in 5 cm steps

150 100 50 50 100 150 z / cm 0.0 2.5 5.0 7.5 10.0 12.5 kHz rate: 76.2 kHz

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 13/ 14

slide-22
SLIDE 22

Conclusion

Background

  • 2 background types: luminosity background (generated at the IP) and

machine background (generated at the walls of the beam pipe)

  • scattering of background tracks at material leads to spread in z
  • ≈ 82 kHz reducible background (tracks not from the IP) and

≈ 40 kHz irreducible background (tracks from the IP)

Neural Network Trigger

  • robust z-vertex estimation with the neural networks
  • signifjcant background reduction with z cut
  • allows to consider a single track trigger

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 14/ 14

slide-23
SLIDE 23

Backup

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 15/ 14

slide-24
SLIDE 24

IP Effjciency

  • effjciency ε:

effjciency to correctly fmag tracks from the IP

  • fake rate FR:

rate of tracks wrongly fmagged as IP tracks

  • split background data in
  • “ip-tracks”: z ∈ [−1, 1] cm
  • “displaced”: z /

∈ [−1, 1] cm

  • vary zcut in 1 cm steps

(zcut ∈ [1..50])

10 20 30 40 fake rate (FR) / kHz 40 60 80 100 efficiency ( ) / %

zcut/cm FR /kHz ε / % 9 5.6 89.5 16 10.9 98.5 22 15.6 99.3 28 20.3 99.9

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 16/ 14

slide-25
SLIDE 25

Background - Suppression

Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

Multiplicity

1 2 3 tracks 20 40 60 80 100 120 kHz

Single Track

10 20 30 40 50 z cut / cm 20 40 60 80 100 120 kHz

Multi Track

10 20 30 40 50 z cut / cm 20 40 60 80 100 120 kHz

rate [kHz] = 1 track ≥ 2 tracks

  • lumi. bkg

89.7 14.5 machine bkg 16.1 1.2 total 105.8 15.7

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 17/ 14

slide-26
SLIDE 26

Introduction - The Belle II Detector

e− e+ Pixel Detector Silicon Vertex Detector

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 18/ 14

slide-27
SLIDE 27

Introduction - The Belle II Detector

e− e+ Central Drift Chamber 56 layers Input for L1 Track Trigger

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 18/ 14

slide-28
SLIDE 28

Introduction - The Belle II Detector

e− e+ Particle Identifjcation Calorimeter KL and µ detector

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 18/ 14

slide-29
SLIDE 29

TwoPhoton

t channel s channel

e+ e+ e− e− e+ e− γ γ e+ e+ e− e− e+ e− e+ e+ e− e− e+ e− e+ e− e+ e− e+ e− Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 19/ 14

slide-30
SLIDE 30

Luminosity Background

  • primary vertex at the IP (z = 0)
  • e+e− from the IP directly hit the CDC
  • back scattered particles hit the CDC

BhabhaM BhabhaL BhabhaS TwoPhoton

pT

1 2 3 4 5 6 pT / GeV 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 kHz rate: 104.2 kHz

z

150 100 50 50 100 150 z / cm 5 10 15 20 25 kHz rate: 104.2 kHz

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 20/ 14

slide-31
SLIDE 31

Machine Background

  • Touschek increase via nano beam scheme
  • small beam pipe (r ≈ 1 cm), resulting in worse vacuum conditions
  • beam gas scattering increased via bad vacuum in the beam pipe

Touschek Coulomb Brems

z

150 100 50 50 100 150 z / cm 0.0 0.5 1.0 1.5 2.0 2.5 3.0 kHz rate: 17.3 kHz

pT

1 2 3 4 5 6 pT / GeV 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 kHz rate: 17.3 kHz

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 21/ 14

slide-32
SLIDE 32

Background - Track Properties

  • ≈ 106 kHz single track background
  • ≈ 16 kHz multi track background
  • most scattered particles: protons (from nuclear spallation)

Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

Particles

e + e

+

p

  • ther

particles 20 40 60 kHz

Multiplicity

1 2 3 tracks 20 40 60 80 100 120 kHz

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 22/ 14

slide-33
SLIDE 33

2 4 6 pT / GeV 1 2 3 4 kHz

rate: 22.0 kHz Touschek Brems BhabhaM BhabhaL BhabhaS TwoPhoton

2 2 phi / rad 0.0 0.5 1.0 1.5 2.0 kHz

rate: 22.0 kHz Touschek Brems BhabhaM BhabhaL BhabhaS TwoPhoton

1.0 0.5 0.0 0.5 1.0 cos(theta) 1 2 3 4 kHz

rate: 22.0 kHz Touschek Brems BhabhaM BhabhaL BhabhaS TwoPhoton

200 100 100 200 z / cm 0.0 2.5 5.0 7.5 10.0 kHz

rate: 22.0 kHz Touschek Brems BhabhaM BhabhaL BhabhaS TwoPhoton

final state e

MC particles after the detector simulation matched to 2D trigger tracks Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 23/ 14

slide-34
SLIDE 34

2 4 6 pT / GeV 1 2 3 kHz

rate: 18.2 kHz Touschek Coulomb BhabhaM BhabhaL BhabhaS TwoPhoton

2 2 phi / rad 0.0 0.5 1.0 1.5 kHz

rate: 18.2 kHz Touschek Coulomb BhabhaM BhabhaL BhabhaS TwoPhoton

1.0 0.5 0.0 0.5 1.0 cos(theta) 0.0 0.5 1.0 1.5 kHz

rate: 18.2 kHz Touschek Coulomb BhabhaM BhabhaL BhabhaS TwoPhoton

200 100 100 200 z / cm 2 4 6 8 10 kHz

rate: 18.2 kHz Touschek Coulomb BhabhaM BhabhaL BhabhaS TwoPhoton

final state e +

MC particles after the detector simulation matched to 2D trigger tracks Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 24/ 14

slide-35
SLIDE 35

0.0 0.5 1.0 1.5 2.0 pT / GeV 0.0 0.2 0.4 0.6 0.8 kHz

rate: 4.8 kHz Touschek Coulomb Brems BhabhaM BhabhaL

2 2 phi / rad 0.00 0.05 0.10 0.15 0.20 kHz

rate: 4.8 kHz Touschek Coulomb Brems BhabhaM BhabhaL

1.0 0.5 0.0 0.5 1.0 cos(theta) 0.0 0.1 0.2 0.3 kHz

rate: 4.8 kHz Touschek Coulomb Brems BhabhaM BhabhaL

200 100 100 200 z / cm 0.0 0.2 0.4 0.6 0.8 kHz

rate: 4.8 kHz Touschek Coulomb Brems BhabhaM BhabhaL

final state

MC particles after the detector simulation matched to 2D trigger tracks Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 25/ 14

slide-36
SLIDE 36

0.0 0.5 1.0 1.5 2.0 pT / GeV 0.0 0.2 0.4 0.6 kHz

rate: 3.7 kHz Touschek Coulomb Brems BhabhaM BhabhaL

2 2 phi / rad 0.00 0.05 0.10 0.15 0.20 kHz

rate: 3.7 kHz Touschek Coulomb Brems BhabhaM BhabhaL

1.0 0.5 0.0 0.5 1.0 cos(theta) 0.0 0.1 0.2 0.3 kHz

rate: 3.7 kHz Touschek Coulomb Brems BhabhaM BhabhaL

200 100 100 200 z / cm 0.0 0.2 0.4 0.6 kHz

rate: 3.7 kHz Touschek Coulomb Brems BhabhaM BhabhaL

final state

+

MC particles after the detector simulation matched to 2D trigger tracks Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 26/ 14

slide-37
SLIDE 37

0.0 0.5 1.0 1.5 2.0 pT / GeV 2 4 6 8 kHz

rate: 72.2 kHz Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

2 2 phi / rad 1 2 3 4 kHz

rate: 72.2 kHz Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

1.0 0.5 0.0 0.5 1.0 cos(theta) 1 2 3 kHz

rate: 72.2 kHz Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

200 100 100 200 z / cm 0.0 2.5 5.0 7.5 10.0 12.5 kHz

rate: 72.2 kHz Touschek Coulomb Brems BhabhaM BhabhaL BhabhaS TwoPhoton

final state p

MC particles after the detector simulation matched to 2D trigger tracks Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 27/ 14

slide-38
SLIDE 38

0.0 0.5 1.0 1.5 2.0 pT / GeV 0.00 0.02 0.04 0.06 kHz

rate: 0.2 kHz Touschek Coulomb Brems BhabhaM BhabhaL

2 2 phi / rad 0.00 0.01 0.02 0.03 0.04 0.05 kHz

rate: 0.2 kHz Touschek Coulomb Brems BhabhaM BhabhaL

1.0 0.5 0.0 0.5 1.0 cos(theta) 0.00 0.01 0.02 0.03 0.04 0.05 kHz

rate: 0.2 kHz Touschek Coulomb Brems BhabhaM BhabhaL

200 100 100 200 z / cm 0.00 0.02 0.04 0.06 0.08 kHz

rate: 0.2 kHz Touschek Coulomb Brems BhabhaM BhabhaL

final state other

MC particles after the detector simulation matched to 2D trigger tracks Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 28/ 14

slide-39
SLIDE 39

Generator particles scattering to fjnal states

e + e 0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 kHz e + e 5 10 15 20 kHz e + e 0.00 0.05 0.10 0.15 0.20 kHz e + e 0.0 0.5 1.0 1.5 2.0 2.5 kHz e + e 0.0 0.5 1.0 1.5 2.0 kHz e + e 5 10 15 20 25 30 35 kHz

e+ e−

  • ther

π− π+ p

Background Suppression with the Belle II Neural Network Trigger (Sebastian Skambraks) 29/ 14