panda software trigger
play

PANDA Software Trigger K. Gtzen Dec. 2012 Challenge Events/Data - PowerPoint PPT Presentation

PANDA Software Trigger K. Gtzen Dec. 2012 Challenge Events/Data acquired by DAQ (temporarily buffered) Software Trigger Algorithms Trickle of events stored on disc Required reduction factor: ~1/1000 (all triggers in total) A


  1. PANDA Software Trigger K. Götzen Dec. 2012

  2. Challenge Events/Data acquired by DAQ (temporarily buffered) Software Trigger Algorithms „Trickle“ of events stored on disc • Required reduction factor: ~1/1000 (all triggers in total) A lot of physics channel triggers → even higher reduction factor required • Dec. 2012 K. Götzen - Status Software Trigger 2

  3. Algorithms: Work in progress • Prerequisites: Tracking, PID, Event building • For now: Study of algorithms based on – Physics Book Channels – Charged particles only – Combinatorics (inclusive) – Invariant masses – PID information Toy MC and Full MC ⇒ See Donghee‘s Talk • • Examples with Toy MC from my own studies Dec. 2012 K. Götzen - Status Software Trigger 3

  4. Definition PID Quality • Asymmetric table, containing – selector efficiencies and # 𝑏𝑑𝑑𝑓𝑞𝑢𝑓𝑒 𝑥𝑠𝑝𝑜𝑕 𝑢𝑧𝑞𝑓 𝑞𝑏𝑠𝑢𝑗𝑑𝑚𝑓𝑡 – misID levels; 𝑛𝑗𝑡𝐽𝐸 = #𝑏𝑚𝑚 𝑥𝑠𝑝𝑜𝑕 𝑢𝑧𝑞𝑓 𝑞𝑏𝑠𝑢𝑗𝑑𝑚𝑓𝑡 Particle Type fraction of pions acc. by electron selector e mu pi K p e eff misID misID misID misID Selector mu misID eff misID misID misID fraction of electrons acc. by pion selector pi misID misID eff misID misID K misID misID misID eff misID p misID misID misID misID eff Dec. 2012 K. Götzen - Status Software Trigger 4

  5. Definition PID Quality • Asymmetric table, containing – selector efficiencies and # 𝑏𝑑𝑑𝑓𝑞𝑢𝑓𝑒 𝑥𝑠𝑝𝑜𝑕 𝑢𝑧𝑞𝑓 𝑞𝑏𝑠𝑢𝑗𝑑𝑚𝑓𝑡 – misID levels; 𝑛𝑗𝑡𝐽𝐸 = #𝑏𝑚𝑚 𝑥𝑠𝑝𝑜𝑕 𝑢𝑧𝑞𝑓 𝑞𝑏𝑠𝑢𝑗𝑑𝑚𝑓𝑡 Particle Type fraction of pions acc. by electron selector e mu pi K p e 0,95 0,05 0,05 0,05 0,05 Selector mu 0,05 0,95 0,05 0,05 0,05 fraction of electrons acc. by pion selector pi 0,05 0,05 0,95 0,05 0,05 K 0,05 0,05 0,05 0,95 0,05 p 0,05 0,05 0,05 0,05 0,95 Dec. 2012 K. Götzen - Status Software Trigger 5

  6. Simultaneous Tagging Examples (Toy MC) 6 tagging algo‘s, D + D - @3.77 GeV ( D → K ππ ) , good PID (5% misID) • D 0 → K - π + J/ ψ→ ℓ + ℓ - D + → K - π + π + ... of 0.3% of the DD events due to BR 2 ... φ → K + K - Λ c → pK - π + + → K + K - π + D s Dec. 2012 K. Götzen - Status Software Trigger 6

  7. Simultaneous Tagging Examples D + D - ( D → any ) , good PID (5% misID) • D 0 → K - π + J/ ψ→ ℓ + ℓ - D + → K - π + π + ... of all(!) DD events φ → K + K - Λ c → pK - π + + → K + K - π + D s Dec. 2012 K. Götzen - Status Software Trigger 7

  8. Simultaneous Tagging Examples • DPM @3.77 GeV, good PID (5% misID) D 0 → K - π + J/ ψ→ ℓ + ℓ - D + → K - π + π +  φ → K + K - Λ c → pK - π + + → K + K - π + D s Dec. 2012 K. Götzen - Status Software Trigger 8

  9. Simultaneous Tagging Examples • DPM @3.77 GeV, perfect PID (0% misID, 100% efficiency) D 0 → K - π + J/ ψ→ ℓ + ℓ - D + → K - π + π +  φ → K + K - Λ c → pK - π + + → K + K - π + D s Dec. 2012 K. Götzen - Status Software Trigger 9

  10. Simultaneous Tagging Examples • DPM @ 5.5 GeV, perfect PID (0% misID, 100% efficiency) D 0 → K - π + J/ ψ→ ℓ + ℓ - D + → K - π + π +  φ → K + K - Λ c → pK - π + + → K + K - π + D s Dec. 2012 K. Götzen - Status Software Trigger 10

  11. Background levels from DPM events Background levels (PID misID = 5%) 14,0 12,0 J/psi -> ll Selected Background [%] 10,0 D0 -> K pi D+ -> K pi pi Ds -> K K pi 8,0 phi -> K K Lc -> p K pi 6,0 All 4,0 2,0 0,0 3,500 4,000 4,500 5,000 5,500 6,000 sqrt(s) [GeV] Dec. 2012 K. Götzen - Status Software Trigger 11

  12. Open Issues for ST • Need full list of all interesting channels! (What‘s gone is gone!) • Selection of D, D s , Λ c , ... – Invariant mass cuts seem insufficient to reduce background – Displaced vertices online for c τ ≈ 50 -200 μ m w/o precise IP? • Performance studied with generic background events from DPM – What if DPM is not realistic? • How treat channels, which are indistinguishable from background? • Environment for testing – Toy MC, Full MC, Online Reco Algorithms ? • PID performance, neutrals quality, event builing quality online • Effect of event mixing/merging (higher combinatorics) on selection performance? Dec. 2012 K. Götzen - Status Software Trigger 12

  13. Event mixing and performance • When do events mix in online scenario? • Assumption: Minimum time between events required for separation Overlap probability (sequent events) 100,00 90,00 50 MHz 80,00 20 MHz Probability [%] 70,00 10 MHz 60,00 5 MHz 2 MHz 50,00 1 MHz 40,00 30,00 20,00 10,00 0,00 0,0 10,0 20,0 30,0 40,0 50,0 Required time difference for event separation • What is performance loss due to more combinatoric? Dec. 2012 K. Götzen - Status Software Trigger 13

  14. Event mixing and performance • Determine effect of higher combinatorics due to event merging • Procedure: – Toy MC (DPM; generator level) background events – Merge fraction of sequent events corresponding to P mix – Apply algorithm and determine amount of background feedthrough – Vary PID mis-ID levels (flat) – Vary center-of-mass energy Dec. 2012 K. Götzen - Status Software Trigger 14

  15. Background levels @ 3.77 GeV Background feedthrough @ 3.77 GeV 100,0 90,0 80,0 Background feedthrough [%] 70,0 no PID misID = 20% 60,0 misID = 5% 50,0 misID = 1% 40,0 30,0 20,0 10,0 0,0 0,00 0,10 0,20 0,30 0,40 0,50 Overlap Probability Dec. 2012 K. Götzen - Status Software Trigger 15

  16. Background levels @ 5.5 GeV Background feedthrough @ 5.5 GeV 100,0 90,0 Event mixing doesn‘t 80,0 no PID Background feedthrough [%] seem to be a huge problem... misID = 20% 70,0 misID = 5% 60,0 misID = 1% 50,0 40,0 30,0 20,0 10,0 0,0 0,00 0,10 0,20 0,30 0,40 0,50 Overlap Probability Dec. 2012 K. Götzen - Status Software Trigger 16

  17. Status • Work being done – Toy MC studies; performance of simultaneous algo‘s – Toy/Full MC studies ( → Donghee ) – Influence of PID on background suppression ( → Donghee ) – Event source simulation ( → Mohammad ) – Online Track reco ( → Yutie, Marius, Sean ) • To do – Compile full list of signatures & develop algorithms – Study neutral particles/channels – Open charm/baryon selection with displace vertices – Alternative background generation Dec. 2012 K. Götzen - Status Software Trigger 17

  18. BACKUP

  19. Why Software Trigger at all? • Many benchmark channels (no ‚golden‘ channel) • Channels consist purely/predominantly of hadrons • Signal and background events look quite similar in terms of – Multiplicity tracks/neutrals – kinematic distributions – event shape, ... Many, many, many more background reactions (  10 6 ) • • No ‚simple‘ hardware trigger can cope with that situation • Need sophisticated algorithms with high selectivity • Only possible with online reco + a lot computing power Dec. 2012 K. Götzen - Status Software Trigger 19

  20. Online Reco ↔ Software Trigger Distinguish between: • Development of the selection/trigger algorithms – Selection algorithms based on information available online → Task of Software Trigger • Online reconstruction/event building – Time ordering / Tracking / Clustering / Track-Cluster-PID- Matching / Event building → Should mainly be addressed by Detector/FEE/DAQ people → Will be addressed both in this session Dec. 2012 K. Götzen - Status Software Trigger 20

  21. Assignment of tasks (till now) • Identification of various selection criteria for relevant PANDA physics channel (cut on momenta, cut on masses, PID, fitting, etc...) → Klaus; DONE • Determination of their selection power of different criteria (ideally in terms of signal efficiency/background suppression). → not assigned; addressed partially in algorithm development • Determination of their dependence from detection quality (efficiency, momentum resolution, energy resolution, partial PID informtation → Donghee's; work in progress • Identification of/search for new selection criteria using more basic information → Donghee + Klaus; work in progress • Evaluation to what extent online event building is a requirement (incl. ↔ excl. triggers), implementation when necessary. Depends on level of event mixing. → Klaus; work in progress • Full implementation of time base simulation for all detectors (in particular pattern recognition and reconstruction based on a time sequential digi stream) → done for some detectors → is up to the detector subgroups. • Implementation of identified algorithms on trigger hardware (FPGA, GPU) → TBD • Test of this hardware with simulated PANDA DAQ input. Use time based simulated detector digis to feed into hardware to test function and performance. → Mohammad Dec. 2012 K. Götzen - Status Software Trigger 21

  22. Quick Compare Toy and Full MC • DPM@3.77 GeV, Toy MC, no PID D 0 → K - π + J/ ψ→ ℓ + ℓ - D + → K - π + π + φ → K + K - Λ c → pK - π + + → K + K - π + D s Dec. 2012 K. Götzen - Status Software Trigger 22

Recommend


More recommend