PRS Muon Results & Status
Darin Acosta
CSC Endcap Muon System
2 endcaps
4 stations (disks) in z
540 chambers
0.5 million channels
PRS Meeting, Mar. 7, 2002
3
Darin Acosta
Overview of CSC DAQ
The pulses from the cathode strips are sampled and
stored in an analog memory (switched capacitor
array) every 50 ns. They are also discriminated and
sent into a comparator network for the trigger
è
The pulses from the anode wire groups are
discriminated, and the digital information is read out
with the trigger data
è
Digitization occurs only if there is a L1 accept
and if there was a local charged track (LCT)
segment in the chamber
p LCT is used to select a “region of interest” in
order to reduce the DAQ bandwidth
p 12-bit ADC for digitization (2 byte word)
è
8 or 16 time samples are read out over a region 16
strips wide by 6 layers deep (3 kB or 5.4 kB)
è
PRS Meeting, Mar. 7, 2002
4
Darin Acosta
CSC DAQ Path
1 DAQMB
serves up to
5 CFEBs
≈ 1 CSC
15 DAQMBs
grouped into
20° sector
(ME1–ME4)
Standalone DAQ
~18 rackmounted PCs
15 optical
fibers
CSC DDU
aka “FED”
× 36
DCC
Sector
Processors
send L1
data
PRS Meeting, Mar. 7, 2002
12 optical fibers
DDU
5
SLINK
Darin Acosta
+1
SLINK
6–9
Data
Concentrator
Cards
CSC Data Rates
Assumptions
LCT occupancy is estimated using ORCA, including neutron
background
è Muon content of L1 triggers is assumed to be 50%
è Jet and e/γ triggers are assumed to have a hard scale, and
increased LCT occupancy is estimated from Pythia+ORCA
è Overhead for S-Link64 headers and empty events is
57.6 MB/s @ 100 kHz DAQ
è
Low Lumi (2×1033):
è
è
è
50 kHz DAQ, ME4 staged, 16 time samples, CLCT selection
500 MB/s (600 MB/s with 3× safety factor on neutrons)
Average occupancy is 1.8 segments ⇒ 10kB/event
High Lumi (1034):
è
è
è
100 kHz DAQ, ME4, 8 time samples, ALCT*CLCT selection
1100 MB/s (1300 MB/s with 3× safety factor on neutrons)
Average occupancy is 3.4 segments ⇒ 10kB/event
PRS Meeting, Mar. 7, 2002
6
Darin Acosta
CSC Calibration
Need to pulse anodes and cathodes:
è
è
è
è
è
è
è
Find dead channels (very quick test)
Measure fC-to-ADC count conversion and linearity (cathodes)
Measure cross-talk
Determine comparator and discriminator thresholds
Measure analog and digital noise
Determine offsets to comparators (cathode trigger)
Inject known patterns to check LCT logic
PRS Meeting, Mar. 7, 2002
7
Darin Acosta
Calibration Methods
Cathodes
Inject charge into one channel of each FE board, read out all
96 channels
è Data volume per pulse: 3.5 MB
è Data volume for full calibration: ~300 GB
è Expect a full calibration to take 10–20 minutes
è Can use standalone DAQ system to collect data
è Full calibration constants checked periodically (monthly)
è Tests require no beams
è
Anodes
è
è
Pulse wire group channels
Smaller amount of data than cathodes
p 1 bit versus 2 bytes, since only have discriminator data
PRS Meeting, Mar. 7, 2002
8
Darin Acosta
CSC Special Triggers
Calibration data is not planned to be taken during normal
collision running (i.e. during abort gaps)
Time synchronization necessary during commissioning
Histogram time distribution, look for LHC gap structure
è Significant amount of running during commissioning, but should
be fixed once normal operation begins
è
Accelerator muon triggers (i.e. “tunnel” or “halo” muons)
CSC Track-Finder (and GMT) will have the ability to trigger on
muons traveling parallel to the beam axis during normal running
è Should be useful for in-situ alignment studies of chambers
è Amount of data needed not known, should not greatly perturb
normal data rates
è
Special loose triggers (heavily prescaled)
Single station trigger to measure CSC Track-Finder efficiency
and to collect sample of collision muons for alignment studies
è
PRS Meeting, Mar. 7, 2002
9
Darin Acosta
DT Barrel Muon System
4 stations in radius
5 wheels in z
250 drift chambers
172K channels
PRS Meeting, Mar. 7, 2002
10
Darin Acosta
DT DAQ Path
172,200 channels
250 chambers
5x12 = 60 sectors
1,440
ROB
60
ROS Master
10,960
FEB
16 channels/FEB
Opt.link
5 DDU
120Mb/s
On chamber
1 per sector
“MINICRATE”
(4 chambers)
In 1 ROB: 4
32-channels TDCs
PRS Meeting, Mar. 7, 2002
DAQ
1 per 12 ROSM
(preliminary;
DDU design yet to be finalized)
11
Darin Acosta
Data rate & size (I)
Fluxes:
a) ≤ 1 tracks/cm2/s everywhere in the Barrel (punchthrough, µ)
(=> may give a BTI candidate in L1 processor)
≅ 30 ‘tracks’/ trigger in the detector @ L1A rate = 100 KHz
=> (12+12+12+8)*30 = 44*30 ≅ 1300 hits/trigger
b) ≤ 10Hz/cm2/s physical (uncorrelated) hits
(e± from neutrons)
( => 10 KHz/channel in worst case (MB1);
noise reduced by
Trigger_rate*TDC_window
< 5% at L1A rate =100KHz ;
a+b)
(including an400
estimated
30%
ns
overhead in headers-trailers )
tdrift(ns)
≅ 1 % channel fired/trigger
4 bytes/channel
PRS Meeting, Mar. 7, 2002
12
tMAX(ns)
Darin Acosta
Data rate & size (II)
(including an estimated 35%
overhead in headers-trailers;
this is in a specific ROS format
proposal; final one yet to be defined)
Total size: 9 KB/event
Small amount of data => all DT data transferred to DAQ @ each L1A
L2 input @ 100 KHz:
900 MB/s to DAQ
=> 900 MB/s / 60 = 120 Mb/s bandwidth on ROS-DDU links
PRS Meeting, Mar. 7, 2002
13
Darin Acosta
RPC Muon System
è
è
è
è
è
è
Six barrel stations in radius
Four endcap stations in z
612 total chambers
160K channels
Digital output
~0.6 kB event size (3% of CSC+DT)
mostly from noise hits
Barrel sector
Endcap sector
PRS Meeting, Mar. 7, 2002
17
Darin Acosta
C M S
R P C M u o n T r ig g e r
R P C M u o n T r ig g e r - s y n c h r o n iz a tio n , c a lib r a tio n a r c h ite c tu r e
L in k B o a r d s
S p litte r s
T r ig g e r
S o r tin g tr e e
F E B o a rd s
R P C
C h a m b e rs
P a c M a tr ix
S o r tin g T r e e
G h o s t B u s te r E ta
G lo b a l
M u o n
T r ig g e r
G h o s t B u s te r P h i
L d e m u x 's
H a r d w a r e s e lf-s y n c h r o n iz a tio n
S y n c h r o n iz a tio n o f te s t p u ls e &
C M S W e e k - P r s /M u M e e tin g - M a r c h 5 , 2 0 0 2
L H C
b u n c h s tru c tu re
I g n a c y M a c ie k K u d la , W a r s a w U n iv e r s ity
CPU analysis of L2-L3 HLT muon
L2 / L3 CPU Analysis:
I Dataset used: W ! 1 + X , full PU, 5000 ev used
I L2 software is official (Muon/MuonTrackFinder &
MuonReco, ORCA 5 3 4) plus navigation optimization:
– in short only the chamber reacheable within FTS error are used.
– not yet in ORCA release, will be soon.
–
5 7 times faster w.r.t. “released” L2.
I L3 “official”;
I hardware: AMD Athlon
1526MP: 1800+
MHz, 512 M B ram
(it’s approx 33% faster than a PIII 1000: MHz)
I Time measurement from TimingReport class by Vincenzo.
S. Lacaprara , PRS/muon meeting , 05-Mar-2002
CPU analysis of L2-L3 HLT muon
L2 CPU analysis (I):
I L2: 780 ms/ev large fluctutation (see after)
– Seed generation 25 ms/ev
– Trajectory builder: 680 ms/ev
– Vertex constraint 75 ms/ev
I Trajectory builder: 680 ms/ev
– Forward K. filter (FTSRefiner): 460 ms/ev (bigger initial error)
– Backward K. filter: 220 ms/ev
of which:
Extrapolation inside DT/CSC chamber: 30 ms/ev
Kalman update: 20 ms/ev
Segment building: 60 ms/ev
S. Lacaprara , PRS/muon meeting , 05-Mar-2002
CPU analysis of L2-L3 HLT muon
L2 CPU analysis (III) (based on
L2all:pt {L2all<2 && pt<50}
htemp
10
Nent = 203
Mean = 0.6579
RMS = 1.102
L2 CPU Time (ms)
# events
L2all
2
200 ev):
2
1.8
1.6
1.4
1.2
10
1
0.8
0.6
1
0.4
0.2
0
2
4
6
8
10
12
L2 CPU time (s)
0
0
10
L2 CPU fluctuation
ptgen = 1:7 GeV, = 1:07, ptL1 = 0!
20
30
40
L2 time vs. pt
slower for low pT
S. Lacaprara , PRS/muon meeting , 05-Mar-2002
50
Pt (GeV)
CPU analysis of L2-L3 HLT muon
L2 CPU analysis (IV):
L2all:eta {L2all<2}
L2all:Gall
L2 CPU time (s)
Geane CPU time (s)
2
12
1.8
1.6
10
1.4
8
1.2
1
6
0.8
4
0.6
0.4
2
0
0
0.2
0
2
4
6
8
10
12
L2 CPU time (s)
all L2 time is spent inside Geane!
-2
-1
0
1
slower in the endcap
S. Lacaprara , PRS/muon meeting , 05-Mar-2002
2
Eta
CPU analysis of L2-L3 HLT muon
L3 CPU analysis:
htemp
L3
L3:
RMS = 1.575
10
I Seed generation 90 ms/ev
I Trajectory builder: 1580 ms/ev
I Trajectory smoother: 8 ms/ev
I no correlation, slower for low
pT muons.
Mean = 1.11
# events
1680 ms/ev
large fluctuation (see ))
Nent = 203
1
0
1
2
S. Lacaprara , PRS/muon meeting , 05-Mar-2002
3
4
5
6
7
8
9
L3 CPU time (s)
–4–
✗
✗
✔
Calorimetric Isolation
Loop over towers and crystals (EcalPlusHcalTower used).
HCAL and ECAL ET deposit around dirAlgo stored in separate
structures, merged later in isolator and compared with thresholds
action
CPU (ms/ev)
extractor: reconstruction
120
extractor: loop over towers
and crystals in towers
+5
isolator:
+ 0.5
✕
✕
✖
✖
M. Konecki
PRS/µ meeting, CMS WEEK, 5 March 2002
–5–
✗
✗
✔
Pixel Isolation
PixelReconstruction used. Searches for Pixel lines around
dirSeed. Analyses pT of lines in cones.
In case of staging (2 pixel layers one 1 disk) isolation can be based on number
of lines in cones.
action CPU (ms/ev)
extractor:
60
isolator:
+ 0.7
✕
✕
✖
✖
M. Konecki
PRS/µ meeting, CMS WEEK, 5 March 2002
Monte Carlo Activities
Detailed work by on understanding how to treat pile-up for
weighted di-muon production (H.Sakulin,S.Lacaprara)
Techniques and statistics required for different samples in order
to achieve a statistical uncertainty of ~10% on final rates at high
luminosity
è
Understanding of how to normalize b cross-section in
min bias events (S.Arcelli, A.Fanfani)
L3 rates dominated by b/c → µ
è Previous PRS/Mu production had mismatch because we
normalized to inclusive muon rate rather than each subprocess
(rate of muons from b may have been overestimated)
è Plan to switch to PRS b/τ approach
è
Cross checks using the Big Min Bias production (D.A.)
è
è
Check of weighting procedure
Evolution of muon occupancy with pT
Misalignment studies (Rivero, Matorras, Calderon)
PRS Meeting, Mar. 7, 2002
18
Darin Acosta
Μisalignment Effects on L2 and L3
trigger using ORCA
• Gravity displacement:
Modified ORCA package to perform
misalignments.
Based on existing Tracker ‘misalignment tools’
∆y = 0
∆y = 10 mm
Muon PRS meeting, 5/3/2002
Rivero, Matorras, Calderon
Conclusion
Typical muon event size is ~20 kB per L1A
è
Data format well understood by HW designers
Muon bandwidth is ~2 GB/s into DAQ
Detailed calibration procedures defined
Online L2/L3 algorithms need significant optimization
Replace GEANE with fast propagator
è Try more radical approaches
è
Thanks to many people in PRS/Muon who worked hard
to provide information and perform studies
PRS Meeting, Mar. 7, 2002
18
Darin Acosta