CMS Luminosity
Measurement(s)
Cornell Journal Club
Daniel Marlow
Princeton University
September 30, 2005
Talk Outline
• Goals & General Strategy
• Real time Techniques
– HF
– Pixel Telescopes
– TAN-region Fast Ionization Chamber
• Offline Techniques
– Total cross section measurement (TOTEM)
– W & Z Counting
• Bookkeeping
• Other Issues
Design Goals: General Desirables
• Absolute calibration, based on a known cross
section with a reliably calculated acceptance.
• Temporal stability against gain changes and
other drifts: “countable objects” or self
calibrating signals (e.g., MIP peak).
• Linearity over a large range of luminosities.
• Real time operation independent of full DAQ.
• Redundancy
– There is no perfect method
– Applies to both real time monitoring and to
offline absolute normalization
Design Goals: Specific Issues
• Real time monitoring
– Bunch by bunch (yes)
– Update time: 0.1 s to 1.0 s or slower
• Offline
– Robust logging
– Easy access to luminosity records
– Dynamic range (1028 ~ 1034 cm–2s–1)
• Absolute Calibration
– Target from previous workshops 5% (or
better)
General Strategy
• Use TOTEM* measurement of total cross
section at low luminosity as a reference point.
• Use real time techniques (HF, Pixel Telescopes,
FIC) to extrapolate/interpolate to design
luminosity
• Renormalize at design luminosity using
processes of ~known cross section (e.g., W’s
and Z’s)
*TOTEM is a forward detector that will measure the total p-p
cross section, thus providing a normalization point.
Lumi Basics
• Apologies to the experts!
• Basic equation
N event   Ldt
• We are often interested in the mean
number of interactions per bunch
crossing:
filled BX
 N   mb L


BX
f
2835
f 
 40 MHz
3564
total BX
Zero Counting
• If μ<<1 measuring the luminosity is
straightforward, since the probability of two
events in a single BX is ~μ2 . It is enough
just to count hits.
• For μ~1, one must either be able to
distinguish between single and double
interactions (not generally possible in this
context), or, one must “count zeroes”
p(n;  ) 
n 
 e
 p(0;  )  e
n!
    log[ p(0)]

Zero Counting
• For μ>>1, one starts to run out of zeroes to count.
There is no hard limit, but requiring at least 1%
zeroes seems reasonable.
This corresponds to
   log(. 01)  4.6
General Strategy
Extrapolation of
the total cross
section as
measured by
TOTEM involves
six orders of
magnitude,* over
which the number
of min bias
interactions goes
from ~10-5 to ~25
per BX.
*The extrapolation is
not quite so bad, since
TOTEM running will be
done with fewer filled
bunches.
High L
Zero famine
Zero feast
TOTEM
Importance of Bunch-byBunch Measurements
• Presumably bunch-by-bunch luminosity
information is of interest to the machine
people.
• But it may also be relevant to physics
simulations. In particular, if there are
significant bunch-to-bunch variations in
the luminosity, then the distribution of the
number of underlying interactions will no
longer be Poisson.
Importance of Bunch-byBunch Measurements
• To see the effect, we consider the case where
the bunch-to-bunch variations in the mean
number of expected interactions is Gaussian
distributed.
• The cases considered range from no smearing
(all bunch luminosities the same) to 50% bunchto-bunch variations.
Importance of Bunch-by-Bunch
Measurements
pure Poisson
10% smearing
20% smearing
50% smearing
The “HF”
The simulation
results to be
presented here will
deal exclusively
with the HF.
HF
HF is fast and transverse shower size is small
5mm
HAD (143 cm)
To cope with high radiation levels (>1
Grad accumulated in 10 years) the
active part is Quartz fibers: the
energy measured through the
Cherenkov light generated by shower
particles.
This is the cause of two of the
peculiar features of this calorimeter:
EM (165 cm)
The visible energy is carried by
relativistic particles, i.e. electrons:
the calorimeter is sensitive to the
EM component (p0) of the hadronic
shower. Shower size depends on
Moliere radius not li
Iron calorimeter
Covers 5 > h > 3
Total of 1728 towers, i.e.
2 x 432 towers for EM and HAD The light is generated preferentially
at 45 degrees: light propagation is
h x f segmentation (0.175 x 0.175)
far from ‘usual’ meridian one.
from Greg Snow Dec. 2002
Signals From HF
Iron fiber
calorimeter.
3<η<5
HF
Minimal add’l hardware requirements
•Mezzanine board to tap into HF
data stream and forward bits to a
PC via Ethernet
•Autonomous (mini) DAQ system
to provide “always on” operation
T1 & T2 are
elements of
TOTEM
HF Energy Depositions
The energy
depositions
in single
interactions
are typically
quite sparse.
Simulation details:
• PYTHIA w. diffractive
events added.
• DC04 (GEANT)
• Extract HF depositions
to Rootuple.
Energy Depositions
At design
luminosity,
there are
typically 25
interactions
per BX.
ET Depositions
Diffractive
interactions
An ET threshold of
1 GeV will detect
most interactions.
Total in one
endcap.
Single
interaction
BX’s only.
HF Methods & Simulation
Methods:
• Count minimum bias events at low luminosity
• Count “zeroes” at design luminosity
• Use also linear ET sum, which scales directly with
luminosity.
Simulation Information:
• Recent work by Chris Rogan, Princeton undergrad.
• Thanks to Monika Grothe & Wisconsin group for
providing MC samples with diffractive events included.
• MC details
–
–
–
–
–
Full GEANT & ORCA
0.25 p.e./GeV
QIE FADC scale realistically included
Thermal noise included
Tested performance over a range of assumptions
HF Zero Counting
• Defeat the zero famine at high luminosity by
counting zeroes in a much smaller solid angle.
• There are 864 HF “physical” towers.
• In effect these provide 864 quasi-independent
measurements of the luminosity.
• Average to arrive at final result.
MC Results: Physical Tower
Zero Counting
<N>/Nexpected
<N>/BX
Zero starvation
3 FADC count
hit threshold
Line is determined
by value at 1034 ,
not a fit.
Deviation from
linearity
<N>/Nexpected
<N>/BX
Physical Tower Zero Counting
w. Increased Threshold
Thresholds ~2X
previous plot
Superlinearity
typical of high
thresholds
<ET>/ETexpected
<ET>/BX (GeV)
MC Results: ET Sum Method
Line is
determined by
value at 1034 ,
not a fit.
This is the average ET
summed over the HF for
each BX. Nominal noise and
threshold (ADC least count
effects included).
Effect of noise at
low luminosity.
Deviation from
linearity
<ET>/ETexpected
<ET>/BX (GeV)
MC Results: ET Sum Method
Noise
Subtracted
Deviation from
linearity
HF Luminosity Readout Path
• 9 HTRs for HF+ and HF• Each HTR has 1 output with luminosity info
– 100Mbps raw ethernet packets sent to router
• Router to computer over Gigabit ethernet
• Dead time, throttle, etc. info from GCT sent to CPU
• This computer will feed LHC, luminosity DB, etc.
HF

H
T
R
H
T
R
R
O
U
T
E
R
9 HTRs/VME
crate
HF
+
H
T
R
Global Trigger
H
T
R
CPU
Luminosity
consumers
Maryland/Princeton/Virginia
HF Luminosity Card
• Initial prototype board built. Next version due
back
TO
P
soon.
– 32MBytes of SDRAM
– Virtex2PRO/VP7 has 722kbits block ram
– Embedded processor
• Function:
– Receive data @ 40MHz from each Xilinx
– Keep running sum of tower occupancy per bucket
– Keep ethernet blasts in memory for transmission
BO
T
Pixel Luminosity Telescope (PLT)
• The HF method is based on an existing
detector, and thus has the advantage of being
inexpensive and relatively easy to implement.
• It does not, however, really fit the bill when it
comes to providing a luminosity measurement
based on “countable objects.”
• Motivated by the CDF approach of counting
MIPs using Cherenkov telescopes, we are
proposing a charged-particle telescope system
based on single crystal diamond detectors.
• This system is not yet fully funded.
Pixel Luminosity Telescope (PLT)
Measure luminosity bunch-by-bunch
• Small angle (~1o) pointing telescopes
• Three planes of diamond sensors (8 mm x 8 mm)
• Diamond pixels bump bonded to CMS pixel ROC
• Form 3-fold coincidence from ROC fast out signal
• Located at
r = 4.5 cm,
• Total length 20 cm
• Eight telescopes per side
Count 3-fold coincidences
on bunch-by-bunch basis.
Simple, stand alone
detector, operating
independently of main
CMS DAQ
z = 175 cm
Rutgers/Princeton/UC Davis
Single Crystal (SC) CVD Diamond
• Radiation hard (few x 1015 cm-2)
• No need for cooling
• Full charge collection at 0.2 V / m
― 18,000 e― for 500 m diamond
― Landau 60% narrower than Si
• Availability of 8 mm x 8mm pieces
― 2 pieces end of last year
― 8 pieces in May
― ready for production by Fall
• Test performance of irradiated
diamond pixel detectors
― efficiency
― spatial resolution
― radiation hardness
Fall Fermilab test beam
Polycrystalline Diamond
Single Crystal:
• distribution well
separated from zero
Polycrystalline:
crystal boundaries
• wide distribution
• low pulse height tail
• charge spreading
over several pixels
very difficult to achieve 1% stability
of efficiency using polycrystaline sensors
CMS Pixel Readout Chip
CMS pixel chip has
“fast” multiplicity
counting built in
80 x 52 pixels
100 m x 150 m
active area
8 mm
Fast output level (each bunch crossing)
• 0, 1, 2, 3,  4 double column hits
• individual pixel thresholds adjustable
• individual pixels can be masked
8 mm
Full pixel readout (every L1 trigger)
• pixel address and pulse height of each hit
• diagnostic of fast out signal
• determination of track origin
• determination of IP location
IP* σxy~35 μm σz~700 μm
Scattering
Beam halo
*statistical accuracy on relative position assuming 1033 for 1 s
Location
• End of Be section of beam pipe (~ 1.7 m from IP)
• Just outside of beam pipe (~ 4 cm from beam line)
IP
Luminosity
telescopes
Rates
for L = 1033
Pythia 6.227 -- CMS Min. Bias productiin
• interactions per BC: 2.5
• number of buckets per orbit: 3564
• filled buckets per orbit: 2835
• tracks per telescope per BC: 0.053
• tracks per array per BC: 0.84
• tracks per telescope per 1 s: 1.7 x 106
• tracks per array per 1 s: 2.7 x 107
• tracks per array per bucket per 1 s: 9500
1.0% statistical luminosity accuracy
per bucket per 1 s @ L = 1033
• ratio of coin./tracks for 18 int. per BC:
0.81
Backgrounds
increase when beam pipe and CMS detector material put in
Coincidences:
11%
almost all coming from photons from
pi0’s interacting in beam pipe
Singles rate:
77%
sources mostly from beam pipe
some from pixel detector
accidental hits per plane per interaction:
0.02
24% fakes in 3-fold coincidences @ L =1034
(assuming accidentals not correlated)
Sources of Hits
• hadrons from physics (IP)
hadron coincidences
• electrons from physics (p0 photon
conversions in beam pipe)
• singles from IP, beam pipe,
pixel detector, telescope planes
electron coincidences
singles
8 telescopes per side φ=0
Top and Bottom of the 3rd plane reduced by
0.5mm
8 telescopes per side φ=22.5
Top and Bottom of the 3rd plane reduced by
0.5mm
4 telescopes per side φ=45
Top and Bottom of the 3rd plane reduced by
1mm
PLT Status
• The PLT is an ideal real-time luminosity monitor.
• A detailed design showing the PLT to be
compatible with CMS has been worked out,
simulation studies are well advanced, and work on
verifying the single-crystal diamonds is in
progress.
• The required funding of ~$300K has not been
secured, but we are pursuing various avenues and
remain optimistic.
• The BCM mechanical mount will accommodate
the PLT.
LHC Luminosity Monitor
• The LHC accelerator project* plans to
incorporate fast ionization counters
(FICs) in the TAN region, which is
±140m from the IP.
TAN
TAN
D1
triplet TAS
TAS
triplet
D1
L n
R
140 m
IP
140 m
*US LARP project in particular.
LHC Luminosity Monitor
Instrumented
Cu bar
absorber
to IP
FIC, which is
integrated
with the ZDC,
is located at
the position
of the beam
septum.
LHC Luminosity Monitor
Target specifications:
• <0.5% relative precision
• Long term stability (~1
month) for calibration with
detectors
• High radiation environment
(100 MGy/year)
• Bunch-by-bunch capability
Solution
• Segmented, multi-gap,
pressurized ArN2 gas
ionization chamber
constructed of rad-hard
materials
Quadrant segmentation
provides sensitivity to
beam position.
LHC Luminosity Monitor
The drift speed in the FIC is such that resolving
individual bunches is an issue. The strategy
being pursued consists of analog pulse shaping
combined with digital deconvolution.
Step 25k
1
1 to 10
2 to 10
3 to 10
4 to 10
5 to 10
6 to 10
7 to 10
8 to 10
9 to 10
10
5
mVolts
4
3
2
1
0
0.8
Channel output [V]
6
0.6
Out @ 20 PP int + 1 PP int
0.4
0.2
0
Out @ 20 PP int
0
50
100
150
nsec
200
250
300
350
-0.2
-20
0
20
40
t [ns]
60
80
100
LHC Luminosity Monitor
•Project is part of LARP, a consortium of
US labs participating in LHC accelerator
work.
•This project is led by LBL.
•The current plan is to provide
information to the LHC machine group.
•I am in contact with the LBL people
and I am working toward getting the
information for incorporation into the
CMS real time luminosity database.
TOTEM
Luminosity
Independent Method
Measure elastic scattering
in Roman Pots and
inelastic in T1 and T2 (see
next slide). Should give
result good to ~1% or so.
Normalization Using W’s and Z’s
M. Dittmar et al.
Basic idea is to use
pp  W   & pp  Z   +  
• Lots of rate
• Well understood
theoretically
• Readily detectable
LHC event rates at 'nominal luminosity'
CMS Trigger TDR
Normalization Using W’s and Z’s
• For a typical process we have
N pp X
2



  partonsX  PDF( x1 , x2 , Q )  Lproton-proton
• If we take pp  Z as a reference process, with
N ppZ  ( q qZ + ho )  PDF ( x1 , x2 , Q 2 )  Lprotonproton
• The ratio does not depend on the luminosity or
the pp cross section.
N ppX  N pp Z 
 q q X
 q qZ
PDF( x1, x2 , Q2 )

PDF( x1 , x2 , Q 2 )
Bookkeeping and Normalization
Consider a process j, whose cross section we wish
to measure
j 
N yield
j
 j L0
yield

 triggered
N
1
1 
j
triggered
N j


N

j
j

 j L0  N triggered
 j L0
j

where N yield
is the event yield in the final sample
j
triggered
(background removed by cuts and fitting), N j
is
the raw number of triggers, εj is the selection
efficiency (trigger plus offline), and L0 is the
luminosity, uncorrected for either deadtime or
prescaling.
A CMS Note on this is in preparation.
. . .General Considerations
Note that
 N yield

j
 j   triggered ,
 Nj

which is the fraction of all events that pass the
analysis cuts, can be determined from any unbiased
sample of triggers. Note that all one needs to know
is the raw number of triggers. Dead times and
prescale factors do not explicitly enter (of course,
one would want to know these as cross checks).
A similar analysis (too complicated for a short talk)
shows that multiple overlapping triggers with
arbitrary prescales can also be accommodated.
Technical Requirements
• A hardware scaler that counts L1A’s before
any prescaling or deadtime losses are applied.
• We need to know the fraction of events
rejected by the HLT.
• This information must be “coordinated” with
the scaler and luminosity information.
Summary and Conclusions
• Various techniques are being pursued for
online luminosity monitoring.
• HF
• PLT
• FIC
• The combination should provide redundancy
and cross checks.
• A simple and robust strategy for obtaining
offline normalization has been formulated.
Extra Slides
ET Depositions
Most of the
energy
depositions in
the physical
towers are
below 1 GeV
Location of IP Centroid
Relative location of IP
• L = 1033
 assume readout rate of 30 kHz
about 1200 tracks per telescope per 1 s
σx = 2.4 mm
• σz = 90 mm
•
• 4 telescopes involved in x (or y) measurement
• 16 telescopes involved in z measurement
@ L = 1033 in 1 s
35 m centroid precision in x (or y)
700 m centroid precision in z
Absolute location of IP
• assume 300 um relative alignment of planes
1.5 mm in x (or y)
30 mm in z
IP Distributions
X
x = 2.4 mm
y = 5.9 mm
z = 89 mm
smeared because of track curvature
and longitudinal beam length (σ~7.5 cm)
Y
Z
What’s Needed
Many parts developed for CMS pixel detector
•
•
•
•
•
•
•
ROC (pixel readout chip)
TBM (readout control chip)
FEC (VME control link module)
FED (VME flash ADC module)
Optical links
Port Card (board for housing lasers, photodiodes,power distribution)
Fan in/out chip (chip for fan out/in of control lines)
A few need to be custom developed
•
•
•
•
•
HDI (circuit on which sensors are mounted)
Bump bonded sensors
Custom chip
Support structure
DAQ board (circuit for histograming of fast out signals)
Bump Bonding
• Bumping of readout chips
 done at PSI as part of pixel project
• Metallization of diamond
 Ti/W pixel electrodes, capability at Rutgers
• Bumping of individual diamond pieces
 UC, Davis has successfully done this in the past
• Flip chipping
 UC, Davis has capability
• Total of 48 pieces need to be bump bonded
• 100 % pixel yield not required
• Diamond sensor can be reworked if bumping fails
TOTEM & CMS
HF
T1
•Need common TOTEM/CMS
running to cross normalize
inelastic cross section.
•Effect of removing T1?
T2 CASTOR