1. The LHCb Level 1 trigger LHC Symposium, October 27, 2001
Frederic Teubert, CERN, EP Division,
on behalf of the LHCb collaboration
The LHCb VELO Detector
The L1 Algorithm and its Performance
New Developments
Outlook
Frederic Teubert

2. VELO Overview Vacuum Vessel Precise vertexing requires, to be:
as close as possible to the decay vertices
with a minimum amount of material between the first measured point and the vertex
 silicon sensors are placed in vacuum
~1 m
Number of silicon sensors: 100 Area of silicon: 0.32 m2
Number of channels: 204,800
Frederic Teubert

3. VELO Setup One detector half: p p forward
Positioning and number of stations is defined
by the LHCb forward angular coverage of
15mrad <  < 390mrad (4.9>>1.6)
together with the inner/outer sensor dimensions
Also need to account for
spread of interaction region: s = 5.3 cm
partial backward coverage for improved primary vertex measurement
One detector half:
Interaction region p p
Final configuration:
25 stations
1 station = 2 modules (left and right)
1 module = 2 sensors (R and )
forward
Frederic Teubert

4. Sensor Design Symmetry of the event production
suggests sensors which measure
 and R coordinates.
Advantages of R geometry:
- Resolution: smallest strip pitch where
it is needed, optimal cost/resolution.
- Radiation: short strips, low noise,
(40mm x 6283mm) close to beam.
- L1:  segmentation of R-sensors (45o)
is enough for primary vertex
reconstruction.
Seeding region
Inner radius is defined by the closest possible approach of any material to the beam: 8 mm (sensitive area)  LHC machine
Outer radius is constrained by the practical wafer size: 42 mm
Frederic Teubert

5. LHCb Trigger System p 40 MHz  1 MHz Fixed latency: 4.0 ms
Data is kept in analog/digital pipelines
L0: look for high pt particles hadron (HCAL), muon (MUON) or electron (ECAL) pile up VETO
40 MHz  1 MHz Fixed latency: 4.0 ms p
L0 YES: Data is digitized, but stays in memory of individual detector electronics
L1: vertex trigger
1 MHz  40 kHz L1 buffer = 1927 events L1 max. latency: 1.7 ms
L1 YES: Digitized data is read-out, event building  CPU farm
L2: refined vertex trigger
40 kHz  5 kHz
5 kHz  O(200 Hz)
L3: online event selection
Frederic Teubert

6. L1 (Vertex) Trigger Input rate: 1 MHz Output rate: 40 kHz
Maximum latency: 1.7 ms
Purpose:
Select events with detached secondary vertices
Needs:
Standalone tracking and vertex reconstruction
Occupancy < 1% per channel
~ 1.5 tracks in the seeding region (innermost 450 sectors of R-sensors)
 track reconstruction is straightforward
Frederic Teubert

7. Algorithm 2d track reconstruction using only R-hits
Efficiency: 98% z
triplets
2d track reconstruction using only R-hits
Primary vertex:
Histogram of two 2d-tracks crossings  z
-segmentation  x,y
sz ~ 70 mm sx,y ~ 20 mm
Frederic Teubert

8. Algorithm 2d tracks Impact Parameter Efficiency: 95% 3d tracks
Minimum bias
Efficiency: 95%
Tracks from PV
3d tracks
(only 2d tracks with large IP)
Search for two 3d tracks close in space (rough secondary vertex)
Minimum bias, fraction of tracks with IP/ > 2
Calculate probability to be a non b-event based on impact parameter and geometry
Frederic Teubert

9. Physics Performance (test beam)
Reconstruction of Primary
Vertex using 2d tracks
(test beam data)
sz ~ 79 mm
Frederic Teubert

10. Execution Time of Algorithm
800 MHz Pentium III
Expect CPU’s to be
10 times faster in 2005
Frederic Teubert

11. Implementation Basic Idea: 150-250 processors SWITCH CPU farm
x20
SWITCH
x100
1 MHz
CPU farm
Readout Unit
Digitizer Board: zero-suppression, common mode correction, cluster finding
Challenge: 4 Gb/s and small event fragments of ~170bytes
2d torus topology based on SCI shared memory
Frederic Teubert

12. Physics Performance (MC studies)
Main limitations:
No momentum information No information on Pt Significance of impact parameter
No particle ID
Working point 40kHz
Frederic Teubert

13. Can we do better?
The L1 topological trigger is mainly limited by the lack of Pt information, so:
Link L0 objects: large Pt (e,,h) with VELO tracks.
Essentially comes “for free”
Get a rough estimation of the momentum for large IP tracks.
Needs a bit of work
Frederic Teubert

14. Level 1 with L0 information
MUON
ECAL
HCAL
B dl = 4 Tm, pt kick = 1.2 GeV/c
matching efficiency
B+ : 78% for at least one 
BJ/()Ks: 96% for at least one 
Frederic Teubert

15. Level 1 with L0 information
BJ/()Ks is saturated using L0().
B+- is improved by a factor >2 due to the Pt information.
Frederic Teubert

16. Level 1 with Pt Maximum Pt Adding Pt information,
Events with at least 2 tracks with IP/ > 3 after L0
  B0s  Ds-(K+K--) K+   B0d  +-
Minimum bias
B0d  +-
Maximum Pt
Adding Pt information,
allows to work with low L1
output rates (5-40) KHz
and high signal efficiencies.
Frederic Teubert

17. Momentum resolution using VELO tracks + TT1:
Level 1 with Pt
Silicon tracking station (TT1),
with ~10% of By
VELO
Momentum resolution using VELO tracks + TT1:
s(1/p) ~ 0.2/p  0.01/GeV
Increase of L1 event size by
~30%
TT1
Frederic Teubert

18. Outlook The feasibility of fast standalone tracking and primary vertex
reconstruction has been checked both with MC and test-beam
data.
The TP L1 algorithm is limited by the lack of Pt information and
lepton ID. The use of L0 information (High Pt: e,,h) improves
the performance.
If a rough estimation of the Pt of the high d0 tracks is available
at L1, one can work with much lower output rates (~10 KHz)
while keeping similar signal efficiencies  Work in progress.
Final answer: Trigger TDR, end 2002
Frederic Teubert