Data acquisi*on and Trigger - Trigger -

Transcription

1 Experimental Methods in Par3cle Physics (HS 2014) Data acquisi*on and Trigger - Trigger - Lea Caminada lea.caminada@physik.uzh.ch 1

2 Interlude: LHC opera3on Data rates at LHC Trigger overview Coincidence logic Pipelines Higher complexity systems Important concepts Occupancy Pileup Spill over Cross talk Random coincidences Dead 3me Overview 2

3 Interlude: LHC opera3on 3

4 LHC 4

5 Proton bunches 2808 bunches per beam protons per bunch 40 MHz bunch collision rate BX: bunch crossing 5

6 LHC Fill An LHC fill usually lasts a few hours AYer injec3on and accelera3on, the beams are first, focused then, brought into collision LHC then declares stable beams à experiments start data- taking Luminosity decreases over the course of the run Divide run into luminosity sec3ons (also called luminosity blocks ) 6

7 Trigger 7

8 Large data volume at the LHC Proton- proton collision rate at LHC is 40 MHz Pixel detector has 66M readout channels out of which O(10 4 ) are hit in an event Data size to be stored for each pixel is 4 byte à Data volume of TB per day! Offline data analysis would take forever Therefore: à Need to reduce data volume online: Trigger à Hardware and/or soyware trigger select events in real 3me based on relevant detector informa3on 8

9 Trigger Concepts In collider experiments, events of interest occurs at much lower rate compared to proton- proton collision rate 9

10 Trigger Strategy Good knowledge of detector and signatures is needed to efficiently select interes3ng events Relevant detector parts and their performance Needed/desired measurement precision Physical proper3es of signal and background events (expected signatures, kinema3c distribu3ons, mass constraints, etc.) In case of mul3purpose experiment: Decide which processes are important Trigger strategy needs detailed planning and has physical, technical and poli3cal aspects. 10

11 Important parameters Timing How long does the trigger need to form decision Need fast processes Need intermediate storage Rate Maximum rate defined by available bandwidth for permanent storage Usually given by background levels Efficiency Efficiency to select signal events Op3mized according to physics needs 11

12 Simple Trigger Setup Trigger setup for a sensor producing a signal at a random 3me (for example cosmic rays, radioac3ve decay) DET DELAY ADC DISCR CONVERSION SIGNAL FOR ADC Discriminator is used to form conversion signal for ADC Signal is passed over delay line 12

13 Coincidence Trigger DET 1 DISCR DELAY SCALER DET 2 DISCR DELAY COINCIDENCE ( AND ) Setup to detect 2- body decay, for example π 0 à ϒϒ 13

14 Coincidence Trigger Input 1 Input 2 Output Coincidence triggers if there is some overlap during 3me window Δt In order to allow for a true coincidence, both signals need to have same length signal path à need to introduce (and adjust) delay lines 14

15 Random Coincides Need to take into account probability that coincidence trigger registers two hits which are not from the same event: P = Δt Z 1 Z 2, where Δt: 3me window of coincidence Z 1, Z 2 : count rate of detector 1,2 TRUE CONINCIDENCDE RANDOM CONINCIDENCDE 15

16 Dead 3me Dead 3me is the 3me that the detector and readout electronics are busy with processing the previous event and not ready to accept new events Any event that happens during the dead 3me is lost! Measures for dead 3me: Total dead 3me d: usually measured in percent as a frac3on of the total measuring 3me Dead 3me per event τ: measured in ms (determined by actual processing 3me in the electronics circuits) 16

17 Dead 3me: Events with random 3me distribu3on Efficiency is related to total dead 3me by: ε = 1 - d For a source with actual event rate R true, the events that can be detected is: R acc = ε R true = (1- d) R true The total dead 3me is related to the dead 3me per event: d = R acc τ = (1- d) R true τ The total dead 3me then becomes: d = (R true τ)/(1+r true τ) And the efficiency becomes: ε = 1/(1+ R true τ) Note that the efficiency decreases with increasing rate Note that there will always be some dead 3me 17

18 Dead 3me: Events with fixed occurrence (collider) Events occur at fixed rate with 3me separa3on t BX Two possible scenario: τ < t BX à no dead 3me (used to be the case at LEP) τ >> t BX à need to keep event rate low! à complex trigger system 18

19 From coincidence triggers to higher complexity Two ways: 1) Larger number of channels à more complex combina3ons - Usually based on FPGAs (field programmable gate array) - Several hundreds inputs - Programmable opera3ons à complex logic combina3ons à Need longer delay lines: 3me for calcula3on increases with complexity 2) Addi3onal computa3ons ayer digi3za3on - For example: π 0 à ϒϒ: Compute π 0 - mass from energy of detected photons, then apply mass window selec3on à Second trigger level (can be built from FPGA or fast processors) à Need long delay lines and intermediate storage 19

20 FPGA Designed to be configured by customer ayer manufacturing using hardware descrip3on language (HDL) Contain large number of logic gates and RAM blocks Logic blocks can be configured to perform complex combinatorial func3ons 20

21 Pipelines Simple delay lines usually not feasible for long delays 100ns delay need about 20m cable Pipelines allow for intermediate storage Analog pipelines: built from switch capacitance Digital pipelines: using digital registers Pipelines consist of several buffer cells. buffer cell 21

22 Pipelines R/W pointers are moved by given clock frequency Chosen to match 3me resolu3on of detector signal buffer cell read pointer write pointer 22

23 Pipelines R/W pointers are moved by given clock frequency Chosen to match 3me resolu3on of detector signal Latency: 3me difference between read and write pointer buffer cell read pointer latency (<buffer length) write pointer 23

24 Pipelines R/W pointers are moved by given clock frequency Chosen to match 3me resolu3on of detector signal Latency: 3me difference between read and write pointer Circular buffer: pointer jump back to first cell when they reach end of buffer buffer cell read pointer latency (<buffer length) write pointer 24

25 Example: Mul3- step trigger for π 0 à ϒϒ Step 1: Digital signal of photon energy from ADC Step 2: Adding the two values Step 3: Comparing the sum to 2 values (upper and lower bound of π 0 - mass window), can be done in parallel Step 4: Store event if m low < m ϒϒ < m high à Trigger latency is 4 t BX à 3 events at the same 3me in the pipeline Note: Trigger decision at LHC takes longer than t BX à trigger decision itself needs to be pipelined 25

26 Occupancy Probability to see a signal in a given channel Aim is to have small occupancy (<<1) Probability to create fake matches (for example hits to tracks) increases with increasing occupancy 26

27 Pileup Electronic pileup: Signal has too long decay 3me à comparator s3ll high when next event arrives à not possible to record new event à data loss 27

28 Pileup Pileup from addi3onal pp collisions at the LHC At LHC, bunches contain protons à several collisions can happen at the same 3me Hard process (the process of interest which triggers the readout) overlayed by par3cles from other collisions Par3cles from secondary collisions will need to be iden3fied offline and subtracted 28

29 Pileup in ATLAS 29

30 Pileup in CMS 30

31 Spill Over At collider experiments, par3cles from previous bunch crossings are s3ll in the detector and seen as signatures for next event Hard to detect on trigger level as whole event informa3on not available Rate can be predicted from simula3on 31

32 CMS Trigger System 32

33 CMS Trigger System 33

34 CMS Trigger System: L1 and HLT LHC BX rate: 40 MHz L1: 100 khz rate 3.2us latency (128 BX) HLT: ~ 300 Hz rate 150ms latency (depends on CPU) 34

35 Fast readout of the detector with limited granularity CMS L1 Trigger Only muon system and calorimeter take part in decision Implementa3on using FPGA and ASICs Synchronous opera3on 35

36 Track segments in the muon system muons Towers of calorimeter cells in ECAL and HCAL: Jets, electrons, photons Total energy, missing energy Isola3on CMS L1 Trigger Objects 36

37 CMS L1 muon trigger 37

38 CMS L1 calo trigger 38

39 CMS L1 Trigger 39

40 CMS L1 Trigger Menu Example Trigger Threshold [GeV] Single muon 16 Double muon 10, 3.5 Isolated double muon 3, 0 Single e/gamma 22 Isolated single e/gamma 20 Double e/gamma 13, 7 Muon + electron 7, 12 Single jet 128 Ouad jet 4 x 40 Six jet 6 x 45 MET 40 HT

41 L1 rate depends on luminosity Trigger menu needs to be adjusted over the course of the experiment 41

42 Trigger prescales Use prescales to adjust rate of a given trigger Needed at high- luminosity runs for some triggers to keep system alive Prescale n: Keep only every n th event (Prescale 1 means no prescale) Dynamic prescales Based on the availability of trigger bandwidth Automa3cally reduce prescales as the luminosity falls over the course of the run Note: Needs to be taken into account in offline physics analysis! 42

43 L1 muon triggers 43

44 L1 jet triggers 44

45 CMS High- Level Trigger (HLT) Events that are accepted by L1 are passed to HLT Full readout of the detector at 100 khz Implemented as soyware algorithms running on large cluster of commercial processors (event filter farm) ~15k cores (30k processors or threads) 45

46 The challenge 46

47 CMS High- Level Trigger (HLT) Form regions of interest to speed up reconstruc3on i.e. if there is a L1 muon or calo tower à perform local reconstruc3on surrounding detectors Reject events as early as possible to free CPUs 47

48 Muons Tracker and muon system Electrons and photons Tracker and calorimeter Taus Tracker and calorimeter Jets, MET, HT Tracker and calorimeter CMS HLT Objects B- tagging (secondary ver3ces) Tracker Other more complex variables 48

49 HLT muons Muons Track segment in muon system Matched to track in tracking detector Isola3on requirement based on calorimeter 49

50 HLT electrons and photons Tower in electromagne3c calorimeter Matched track in tracking detector? yes: electron, no: photon Isola3on requirement based on calorimeter 50

51 HLT taus Leptonic tau decays (e,mu) usign e/mu triggers Hadronic tau decays: 1- prong or 3- prong decays Calo cluster matched to tracks in tracker detector Isola3on requirement based on calorimeter detector 51

52 HLT b- tagging Calo cluster with associated tracks in the tracker detector B- tagging based on long life3me of b- quark and its large mass Form secondary vertex from tracks in jet 52

53 HLT menu 53

54 CMS HLT Trigger Menu Example Trigger Threshold [GeV] Single muon 40 Single isolated muon 24 Double muon 17, 8 Single electron 80 Isolated single electron 27 Single photon 150 Double photon 36, 22 Muon + electron 8, 17 Single jet 320 Ouad jet 4 x 80 Six jet 6 x 45 MET 120 HT

55 HLT rate depends on luminosity Different HLT paths have different allocated rates 55

56 HLT muon efficiency Measured efficiency compared to simulated efficiency Reasonable agreement between data and simula3on Note: Need to have ways of measuring trigger efficiency 56

57 Measurement of trigger efficiency Knowledge of trigger efficiency needed for physics analysis Strategy for measuring trigger efficiency needs to be thought of from the very beginning Events that are rejected on trigger level are lost, i.e. not available for efficiency calcula3on! à need backup trigger with looser selec3on à can be prescaled 57

58 Measurement of trigger efficiency Example: Measurement of efficiency of QuadJet50 Efficiency measured wrt reconstructed events 8 jets p T > 30 GeV, leading 4 jets p T > 50 GeV Use EightJet30 as backup trigger à need to take bias into account Efficiency Data MC MC (HLT_8j30) jet p T [GeV] 58

59 Summary Triggers are needed to reduce data volume and allow for permanent storage Trigger strategy has physical, technical and poli3cal aspects and needs careful planning Parameters to consider are 3ming, rate and signal efficiency Discussed trigger concepts ranging from simple coincidence triggers to mul3- level trigger system as used at the LHC 59