BTeV Level-1 Trigger Segment Finding Algorithm

1
BTeV Level-1 TriggerSegment Finding Algorithm

• Internships for Physics Majors, August 8th, 2002
• Jennifer Seiler, Fermilab
• (supervised by Jin-Yuan Wu)

2
BTeV detector

• Silicon pixel vertex detector provides
• Pattern recognition power
• Very good position resolution (7m)
• Radiation hardness

3
Si- pixel Detector
7.2 mm
8 mm
A pixel

• 50 ?m x 400 ?m pixels bump-bonded to readout
  chips 22 million total
• Two pixel planes per station (supported on a
  single substrate), with 30 planes (a station is
  10 cm tall and they are 4.42 cm apart)
• Beam hole 12 mm x 12 mm
• Detectors in vacuum
• Half planes move together when Tevatron beams are
  stable.

4
BTeV trigger block diagram
5
L1 vertex trigger block diagram
6
Triplet to Track Finding (T2T)
10 cm
7
L1 vertex trigger algorithm
FPGA Segment Finder (Pattern Recognition)

• Find beginning and ending segments of tracks from
  hit clusters in 3 adjacent stations (triplets)
• beginning segments required to originate from
  beam region
• ending segments required to project out of pixel
  detector volume

DSP Tracking and Vertexing

• Match beginning and ending segments found by FPGA
  segment finder to form complete tracks
• Reconstruct primary interaction vertices using
  complete tracks
• Find tracks that are detached from
  reconstructed primaries

8
L1 vertex trigger algorithm
Execute Trigger
9
L1 trigger efficiencies
Process Eff. () Monte Carlo

Minimum bias 1 BTeVGeant
Bs DsK- 74 BTeVGeant
B0 Dr - 64 BTeVGeant
B0 r0p 0 56 BTeVGeant
B0 J/yK s 50 BTeVGeant
Bs J/yK 0 68 MCFast
B- D0K - 70 MCFast
B- Ksp - 27 MCFast
B0 2-body modes 63 MCFast
(pp-,Kp-,KK-)
10
Summary Problems Options
Problems Seen

• False track / Good Track Found ratio .1502 on
  average
• In false tracks the false hit is most often the
  first station hit
• Many of the false tracks we are seeing are too
  close to be resolved, other false tracks
  intersect at small angles with real tracks
• 85 as many external tracks found as internal
  found
• Internal are closer together and harder to
  resolve
• External track finding is very complete and
  accurate

Options

• One fix I did was to change to equation for
  finding matching precision-x hits in the non-bend
  view, so that precision-x hits for a give station
  would be found using only the other two stations
• Another possibility is to find tracks by finding
  short precision doublets 1st
• And a third is to find secondary interior
  triplets and see how they overlap onto the
  primary interior triplets, as a quadruplet
  finding algorithm
• All these options have potential problems

11
Proximity Issues
Too Many Internal Tracks

• 85 as many external as internal tracks
• Means extra data flow and more trouble matching
  triplets to tracks
• Also means lower triggering rate
• Options
• Tighter requirements for interior matches than
  for exterior
• Quadruplet finding by overlapping triplets for
  interior

12
Found versus Real Tracks
Found Tracks

• Few real tracks are missed, but they exist among
  the extra tracks that are found
• Internal tracks find more extra tracks, but all
  of the tracks it finds are basically reasonable
  to consider as valid tracks

13
What BB33 is Finding
Dir Sta x1 y1 z1
x2 y2 z2 x3
y3 z3 track 1 2
3 All -1 5 0.08191 -0.60437 -46.92006
0.06346 -0.65716 -51.17007 0.08923 -0.71165
-55.42007 67 67 67 67 0
5 0.00000 0.00000 0.000000 0.07918 -0.65196
-50.77007 0.08437 -0.70800 -55.02007 -99999
67 67 -1 6 0.17461 0.74860 -42.67004
0.31188 0.93675 -46.92005 0.51947 1.17190
-51.17004 766 484 934 0 0
6 0.16701 0.74222 -42.27004 0.31291 0.93336
-46.52005 0.00000 0.00000 0.00000 766
484 -99999 -1 7 0.15769 0.63582 -38.42004
0.17461 0.74860 -42.67004 0.17878 0.86394
-46.92005 766 766 766 766 0
7 0.14627 0.64118 -38.02003 0.16701 0.74222
-42.27004 0.18857 0.86125 -46.52005 766
766 766 -1 7 0.25261 0.71937 -38.42004
0.28862 0.82680 -42.67004 0.31188 0.93675
-46.92005 484 484 484 484 0
7 0.25893 0.72209 -38.02004 0.28442 0.81960
-42.27004 0.31291 0.93336 -46.52005 484
484 484 -1 11 0.76146 -0.31655 -21.42002
1.04804 -0.38669 -25.67002 1.32104 -0.43427
-29.92003 716 716 716 716 0
11 0.73228 -0.32195 -21.02003 1.00861 -0.38545
-25.27003 1.28699 -0.42141 -29.52003 716
716 716 -1 11 0.79615 -0.36298 -21.42002
1.04804 -0.38669 -25.67002 1.32104 -0.43427
-29.92003 502 716 716 0 0
11 0.79318 -0.34968 -21.02002 1.00861 -0.38545
-25.27003 1.28699 -0.42141 -29.52003 502
716 716 -1 11 0.76146 -0.31655 -21.42002
0.95540 -0.51116 -25.67003 1.11232 -0.68513
-29.92003 716 502 502 0 0
11 0.73228 -0.32195 -21.02003 0.94011 -0.51084
-25.27002 1.08892 -0.67355 -29.52002 716
502 502 -1 11 0.79615 -0.36298 -21.42002
0.95540 -0.51116 -25.67003 1.11232 -0.68513
-29.92003 502 502 502 502 0
11 0.79318 -0.34968 -21.02002 0.94011 -0.51084
-25.27002 1.08892 -0.67355 -29.52002 502
502 502 -1 14 -0.18752 -0.70090 -8.67001
-0.24316 -0.96602 -12.92001 -0.35650 -1.20455
-17.17002 525 525 525 525 0
14 -0.18019 -0.67240 -8.27002 -0.25718 -0.95745
-12.52002 -0.33892 -1.20030 -16.77003 525
525 525 -1 14 -0.06413 -0.74273 -8.67002
-0.24316 -0.96602 -12.92001 -0.35650 -1.20455
-17.17002 78 525 525 0 0
14 -0.03917 -0.72432 -8.27002 -0.25718 -0.95745
-12.52002 -0.33892 -1.20030 -16.77003 78
525 525 -1 14 -0.06413 -0.74273 -8.67002
-0.05126 -0.92096 -12.92003 -0.04046 -1.11612
-17.17003 78 78 78 78 0
14 -0.03917 -0.72432 -8.27002 -0.04237 -0.91233
-12.54720 -0.04744 -1.11445 -16.77003 78
868 78 -1 16 -0.52026 -0.60268 -0.17000
-0.67110 -0.84793 -4.42002 -0.84493 -1.05162
-8.67002 841 247 247 0 0
16 0.00000 0.00000 0.00000 -0.67105 -0.80221
-4.02002 -0.81364 -1.02810 -8.27002 -99999
247 247 -1 16 -0.54949 -0.61751 -0.22999
-0.67110 -0.84793 -4.42002 -0.84493 -1.05162
-8.67002 736 247 247 0 0
16 -0.52883 -0.62656 0.22999 -0.67105 -0.80221
-4.02002 -0.81364 -1.02810 -8.27002 247
247 247 -1 16 -0.53961 -0.65092 -0.17002
-0.67110 -0.84793 -4.42002 -0.84493 -1.05162
-8.67002 247 247 247 247 0
16 -0.52883 -0.62656 0.22999 -0.67105 -0.80221
-4.02002 -0.81364 -1.02810 -8.27002 247
247 247
14
Whats Happening Here?

• Tracks crossing at small angles
• Tracks too close
• Tracks options with different curvatures in the
  bend view
• Third hit issues occur with from issue like the
  one illustrated in the second image (middle hit
  confusion occurs most rarely)

15
Beyond Track Finding Options
Short Doublet Finding
BB33

• Finding tracks by first looking at the short
  doublets and projecting forward can be done one
  many ways
• Projecting to other doublets and comparing slopes
  to make a long doublet, then projecting forward
  again
• Projecting the short doublet to the next plane
  and then the next plane

New
Short Doublet Finding Issues

• A good portion of the good tracks have missing
  precision-x matches
• A small percentage of false tracks can be
  eliminated with missing or bad short doublet hits
  and a significant portion are falsely confirmed
• Lower resolution in either dimension

16
Beyond Track-Finding Issues
Secondary Triplets (Quadruplets)

• Quadruplets can be found without new hardware by
  finding secondary triplets and seeing how they
  overlap onto the primary interior triplets
• Allows for another confirming hit
• Allows for better delta dy and delta dx checks

Secondary Triplets Issues

• Significantly higher data flow (gt20)
• The secondary triplets will only have as many
  false tracks
• Trade-off slightly higher computation time on
  finding more triplets and matching them, but cuts
  computation for matching false interior tracks to
  exterior tracks

17
The False Precision Hit Problem
Solution
False Precision Hits

• The BB33 algorithm finds the short doublet match
  at a station by looking between that station and
  the next
• This finds the doublets for real and false tracks
  alike
• My algorithm projects the other two stations onto
  the station in question (only in the non-bend
  view), then looks at the slope of that short
  doublet

BB33
New
Problem Only Works in non-bend view
18
Quick note
Doublet Finding

• Good to Bad doublet ratio ?13 (reasonable,
  maybe? But could be lower?)
• Short doublet finding might give fewer false
  doublets, but a number of good tracks with
  missing false doublets would be lost, and most
  false tracks (though a lower percentage than real
  tracks) find precision hits
• Any suggestions?

19
Bad versus Good Triplets
False Tracks Hits
BB33 All Triplets
My All Triplets
20
Which Hits Are Bad
BB33
My
21
Missing Precision-x Matches
BB33 Triplets
My Triplets
22
Whats to be done?

• Conclusions
• A short doublet algorithm wouldnt be very
  helpful, because good tracks would be lost and it
  would give an almost equal ratio of badgood
  tracks
• Using the other two stations to find the
  precision hits on for a station in a triplet
  lowers the frequency of false tracks
• A quadruplet algorithm while costly in time and
  bandwidth, could improve track finding efficiency
• We need more ideas
• Further Study
• I have yet to finish my quadruplet finding
  algorithm.
• If I have the time this summer I will attempt a
  short doublet algorithm, if no other ideas are
  available

23
QOTDs May-Aug 2002

• "Turns out that this algorithm doesn't work, but
  it has a worse problem....it's very slow"- Myers
  (CS312 professor)
• Then I stared blankly at the dashboard of the
  car with a look on my face which suggested I was
  trying to calculate in my head the square root of
  oatmeal. Dave Schoonover
• In the hall at Fermilab Yeah, Im taking a
  vacation in August with the family. I think Ill
  head out to SLAC.
• Does anyone else want to smell the projector?
  Nathan
• These observations are actually correlated and
  connected! (Symmetries in K-Physics, 7/9/02)
• In the Cornell Graphics lab what are you
  doing? Are you playing Quake Battlefield?
  UhumnoIm .uhdoing graphics researchyeah.
  the physics engine
• My emacs says, "TeX-shell" and I saw that and
  thought of tacos Jerry, 7/11/02
• You have a beam of electrons going around the
  accelerator. And then they get to the wiggler
  and the wiggler shakes them like this. ltshakes
  handsgt And this pisses the electrons off, so
  they radiate more. Jerry, 7/16/02
• The NSA has a website. And it has a "Kid's
  Page." I sense fundamental wrongness. Chase,
  7/8/02

24
More QOTD 5/02-8/02

• A Friday field trip to explore the possible
  combinations and interactions of Subway
  food-multiplets with the 19-dim
  summer-student-group peopleDue to zthe hogh
  calculation efforts one should consider a pretest
  with a smaller low-dimensional subgroup Thats
  not silly!!! It is supposed to find the
  underlying field structure of the student-student
  interactions transmitted via the food-force. A
  very serious research topic!! Andreas
• The velocity selector is like having a bunch of
  little elves inside the box looking at all the
  particles, and if it's not exactly the right
  speed, they molest it. Dr. Dell
• Einstein always asked the right questions -- the
  ones that had answers. Except for that unified
  field theory thing, but, oh well, he was old.
  Dr. Dell
• Imagine I'm this massless thing with two balls.
  Dr. Dell
• "If you're out in space and have no environment
  to interact with, you can always rip off your arm
  and throw it out. And, because the center of mass
  of the system remains constant, you'll just be
  this bloody mass accelerating backwards. Of
  course, when you got to your spaceship, you'd
  immediately cauterize the wound with some bizarre
  laser device." Dr. Dell
• Q How is a chain saw like a Hilbert space? A
  Both comprise an infinite number of infinitesimal
  axes. Eric 7/29/02

25
Acknowledgements

• I would like to thank my family for allowing me
  to leave home for the 6th summer in a row, and
  for supporting, though discouraging, my love of
  physics.
• I would like to thank my friends for making
  physics that much more fun, geeks that they all
  are.
• Thanks to Michael Wang for arranging the code for
  me, and answering many of my questions.
• Most of all, I would like to thank Eric, Roger,
  Max, and my supervisor, Jin-yuan Wu, for making
  all of this possible.
• And Thanks too all of the Fermilab Interns for
  making this such a fun and wonderful experience.