Transcript L1 and HLT combined trigger network

The Level-0 Calorimeter
Trigger
and the software triggers
Umberto Marconi
INFN Bologna
CSN1
Lecce, 24 September 2003
1
Overview
■
The L0 calorimeter trigger
● The Selection Crate
● Project Cost
● 2004 activity plan
■
The L1&HLT trigger
● LHCb Trigger System
CERN/LHCC 2003-31, LHCb TDR 10, September
2003
● LHCb Reoptimized Detector Design and Performance
CERN/LHCC 2003-30, LHCb TDR 9, September
2003
U. Marconi
INFN, Bologna
2
Trigger Architecture
■
Level-0 trigger (fixed latency
4μs)
● Reduce the 10MHz visible
interaction rate to 1.1 MHz
● Event selection:
▪
▪
■
the highest ET hadron,
electron,photon
The two highest pT muons
Level-1 trigger (variable
latency, 58ms max)
● Output rate is fixed at 40KHz
● The decision is delivered
chronologically ordered
● Event selection: B vertex
■
HLT (variable latency)
● Output rate is established at
200 Hz
● Event selection: algorithms
for specific decay modes
U. Marconi
INFN, Bologna
3
The L0 Trigger
■
■
Hardware trigger, custom electronics
4 systems
● Calorimeter trigger system
▪ Select high ET particles from their calorimeter deposits
▪ Electrons, photons, π0 and hadrons
● Muon trigger system
▪ Select high PT muons with fast tracking
● Pile-Up system
▪ Measures the number of p-p interaction in each crossing
● Decision unit
▪ Combines the information from the 3 sources, propose the
decision to the Readout Supervisor.
U. Marconi
INFN, Bologna
4
L0 Calorimeter Trigger
■
Detect a local high ET cluster in
ECAL or HCAL
● 2x2 cells
▪
From 8x8 cm2 (Inner ECAL) to 52x52
cm2 (Outer HCAL)
● ECAL: 5952 cells -> 5952 possible
clusters
▪
▪
Local Maximum
Validation by PS/SPD (same
geometry) to get electron and photon
candidates
Combination to get π0 candidates.
● HCAL : 1484 cells -> 1484 possible
clusters
▪
U. Marconi
INFN, Bologna
Add the ECAL ET in front if available
Validation
5
The Selection Crate
■
Data input @ 40 MHz
● 28 electron clusters
● 28 photon clusters
● 2x28 neutral pion clusters (local,
global)
● 80 hadron clusters
● 16 SPD hits partial sums
■
Data output to the L0DU
@40MHz
● Highest transverse energy cluster
for each cluster type
(5 highest)
● Total transverse hadron energy
(global trigger variable)
● Total SPD hit multiplicity
(global trigger variable)
■
Data output to the L1 trigger
@1.1MHz
● The entire set of the processed
clusters
U. Marconi
INFN, Bologna
LHCb Note 2003-095
6
The Selection Board
Input Interface
U. Marconi
INFN, Bologna
Processing Unit
Slow Control
Fast Control
Output Interface
7
Test of the prototypes (I)
The Processing Unit functionalities are implemented in a
single FPGA
1150 I/O pins FPGA
U. Marconi
INFN, Bologna
8
Test of Prototypes (II)
The optical link to be used
in the output interface
has been successfully
tested
The level of the BER is around 10-13
even in the worst jitter conditions of
about 10-2ps peak-to-peak
U. Marconi
INFN, Bologna
9
Test of the Prototypes(III)
■
We are ready to start testing the input interface
● The prototype hosts a single 12 channels optical/voltage
transducer and 8 deserializers
● It is built as a 1/3 of a 9U standard board
U. Marconi
INFN, Bologna
10
New Cost Estimate
Old vs New Implementation
■
■
■
The allocated INFN budget to build up the Level-0
Calorimeter Trigger is of 800KCHF
The cost now is of about 450KCHF
2004 plans
● Build up the final prototype integrating the Input Interface, the
Processing Unit, the Ouput Interface and the Slow Control (ECS)
in one 9U board
U. Marconi
INFN, Bologna
Electron
Photon
SPD
Neutral Pion
Hadron
Old Impl.
#Boards
2
2
0
4
8
New Impl.
#Boards
1
1
1
2
3
11
Sharing the L0 Bandwidth
■
The L0 thresholds has been defined by requiring the bandwidth
division among a set of reference channels that minimizes the overall
loss in efficiency, by miximizing
Σch (εL0ch / εL0ch-max)
with the constraint of 1.1MHz ouput rate
where
● εch-max is the efficiency when the full bandwidth is available
● εch is the efficiency obtained for a set of thresholds
U. Marconi
INFN, Bologna
Cut (GeV)
M.B. Rate
(KHz)
Hadron
ET =3.6
705
Electron
ET = 2.8
103
Photon
ET = 2.6
126
π° local
ET = 4.5
110
π° global
ET = 4.0
145
muon
pT = 1.1
110
ΣpμT
pT = 1.3
145
12
L0 Trigger Efficiencies vs
L0 output rate
εL0ch
ε L0ch-max
Full bandwidth
available to the
channel
U. Marconi
INFN, Bologna
13
L0 Trigger Efficiency
for Offline Selected Signal Channels
HCAL Trigger
dominates
U. Marconi
INFN, Bologna
MuonTrigger
Electron Trigger
dominates
dominates
14
L0 Trigger Efficiency
for Offline Selected Signal Channels
Trigger Correlations
Decay Channel
εL0/sel (%)
Hadron
Trigger
Electron
Trigger
Muon
Trigger
B° π+π-
53.6
47.6
14.1
6.8
B° π+π- π°
77.2
39.4
66.2
7.9
B° D*-π+
49.0
41.7
14.0
8.4
B°s Ds-π+
49.4
42.2
13.1
8.3
B°s Ds-K+
47.2
39.4
11.7
8.2
B° J/Ψ(μμ)K°s
89.3
18.6
8.3
87.2
B° J/Ψ(ee)K°s
48.3
21.5
37.4
7.0
B°s J/Ψ(μμ)Φ(K+K-)
89.7
20.0
8.4
87.4
B° K*(K+ π-)
72.9
32.7
68.1
7.8
U. Marconi
INFN, Bologna
15
Tracks
T track
Upstream track
VELO
seeds
VELO track
Long track (forward)
Long track (matched)
T seeds
Downstream track
Long tracks
 highest quality for physics (good IP & p resolution)
Downstream tracks  needed for efficient KS finding (good p resolution)
Upstream tracks
 lower p, worse p resolution, but useful for RICH1 pattern recognition
T tracks
 useful for RICH2 pattern recognition
VELO tracks
 useful for primary vertex reconstruction (good IP resolution)
U. Marconi
INFN, Bologna
16
L1 Trigger
2D RZ-tracks in the VELO
PV reconstruction
3D reconstruction (of large
impact parameter tracks)
Measure pT using the 3D VELO tracks to TT
then 3D matching to the L0 objects
RMSx,y=25μm
RMSz =60μm
250cm
U. Marconi
INFN, Bologna
17
VELO-TT Tracking in L1
In average 8.5 tracks need to be reconstructed in 3D
from 60 forward 2D tracks in the VELO in a minimum
bias event requiring an impact parameter between 0.15
and 3 mm
U. Marconi
INFN, Bologna
The reconstruction efficiency on reconstructable B-decay
product is 94%
18
The Level-1 Trigger Variables
Mbias
Mbias
Bππ
U. Marconi
INFN, Bologna
Mbias
19
Level-1 Efficiencies
U. Marconi
INFN, Bologna
Decay Channel
εL1 (%)
B° π+π-
62.7
B° π+π- π°
46.6
B° D*-π+
56.0
B°s Ds-π+
63.0
B°s Ds-K+
62.6
B° J/Ψ(μμ)K°s
67.7
B° J/Ψ(ee)K°s
54.9
B°s J/Ψ(μμ)Φ(K+K-)
71.4
B° K*(K+ π-)
51.9
20
L1 Computing Time
On average 8.3ms is spent in L1 with
the algorithm on a 1GHz PIII CPU
(not optimized to allow large scale
stabilized physics studies)
1 GHz Pentium III
The L1 trigger algorithm has been
speed up in the meanwhile. The
average processing time is now 6ms
The average processing time per
event is expected will be less than
1ms in the 2007
Present time+Moore’s law: <1000
CPU nodes
U. Marconi
INFN, Bologna
21
HLT L1-Confirmation
HLT (L1 Confirmation)
10MHz of visible pp interaction
contains 100KHz of beauty events
bb(%)
cc(%)
Generated
1.1
5.6
Level-0
3.0
10.6
Level-1
9.7
14.2
HLT-L1C
14.0
14.7
U. Marconi
INFN, Bologna
The beauty content of the 20 KHz of the events
which have to be analyzed after the HLT L1-confirmation
is still dominated by light quark events
22
HLT Selection Algorithms
■
■
■
The combined output rate of all
channels under study at the
moment in LHCb, including the
background, is less than 1Hz
The HLT will have a
significatively larger output rate
due the need to relax the final
selection cuts to study the
sensitivity and systematics
The RICH reconstruction on
the event passing the HLT filter
can be executed at a rate of a
few hundred Hz
U. Marconi
INFN, Bologna
Algorithm
HLT rate (Hz)
B J/Ψ(μμ)X
21±4
B h+h-
12±3
B Dsh
12±3
B° K*0(K+ π-)π+
13±3
23
Efficiencies, Event Yields and
Bbb/S ratios
Det.
eff.
(%)
B0   
12.2
Bs  K K
12.0
Bs  Ds 
5.4
Bs  Ds+ K
5.4
B0  D~0 (K)K*0
5.3
B0  J/() K0S
6.5
B0  J/(ee) K0S
5.8
Bs  J/() 
7.6
Bs  J/(ee) 
6.7
B0  
6.0
B0  K* 
9.5
Bs   
9.7
+ few more channe ls in T DR
U. Marconi
INFN, Bologna
Rec.
eff.
(%)
91.6
92.5
80.6
82.0
81.8
66.5
60.8
82.5
76.5
65.5
86.8
86.3
Sel.
eff.
(%)
18.3
28.6
25.0
20.6
22.9
53.5
17.7
41.6
22.0
2.0
5.0
7.6
Trig.
eff.
(%)
33.6
36.7
31.1
29.5
35.4
60.5
26.5
64.0
28.0
36.0
37.8
34.3
Tot.
eff.
(%)
0.69
0.99
0.34
0.27
0.35
1.39
0.16
1.67
0.32
0.03
0.16
0.22
Vis. Annual
BR
signal
6
(10 )
yield
4.8
26k
18.5
37k
120.
80k
10.
5.4k
1.2
3.4k
20.
216k
20.
26k
31.
100k
31.
20k
20.
4.4k
29.
35k
21.
9.3k
B/S
from
bb bkg.
< 0.7
0.3
0.3
< 1.0
< 0.5
0.8
1.0
< 0.3
0.7
< 7.1
< 0.7
< 2.4
Nominal year = 1012 bb pairs produced (107 s at L=21032 cm2s1 with bb=500 b)
Yields include factor 2 from CP-conjugated decays
24
Branching ratios from PDG or SM predictions
Conclusions
■
The design of the Selection Crate is almost
complete.
● The 28 input channels optical interface of the
Selection Board still has to be tested
■
■
■
A final prototype of the Selection Board will be
built in the 2004
The design phase of the L1&HLT triggers of
LHCb is finished.
The trigger system is flexible, robust, efficient.
U. Marconi
INFN, Bologna
25
Monte Carlo Generation
■
pp interactions
● Minimum bias events from PYTHIA 6.2
total = 100 mb
visible = 65 mb
▪ Hard QCD processes, single and double diffraction
▪ Multiple parton interactions tuned to reproduced track
multiplicities observed at SPS and Tevatron energies
● bb events
▪ Extracted from minimum bias sample
■
bb/ visible = 0.8%
bunch crossings in LHCb
x= y = 70 m, z = 5 cm
● Size of luminous region
● Simultaneous pp interactions (“pileup”)
▪ number of visible interactions n (in events with at least one)
distributed according to L = 2  1032 cm2 s1, <> = 30 MHz
U. Marconi
INFN, Bologna
At least two tracks
reconstructible in
whole spectrometer
<n>min bias = 1.23
<n>bb event = 1.42
26
Result of Track Finding
T1 T2 T3
Typical event display:
Red = measurements (hits)
Blue = all reconstructed tracks
TT
VELO
Average multiplicity in bb event
26 long tracks
4 downstream tracks
11 upstream tracks
5 T tracks
26 VELO tracks
U. Marconi=
Total
72 tracks
INFN, Bologna
<p/p>
0.37%
0.43%
~15%
efficiency
94% for p >10 GeV/c
80% for p > 5 GeV/c
75% for p > 1 GeV/c
(IP)
40 m
2050 hits assigned to a long track
98.7% correctly assigned
27