This presentation is the property of its rightful owner.
1 / 16

# Azimuthal symmetry of HCAL readouts PowerPoint PPT Presentation

Azimuthal symmetry of HCAL readouts. O. Kodolova (SINP MSU). f symmetry: goal and data stream choice. Goal is to install the relative scale within each eta-ring. - using the phi-symmetric events equalize the response of HCAL readouts

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

#### Presentation Transcript

O. Kodolova (SINP MSU)

f symmetry: goal and data stream choice

Goal is to install the relative scale within each eta-ring.

- using the phi-symmetric events equalize the response of HCAL

- choice of data stream may be different for the different eta

regions: HB/HE/HF

Current data stream choices:

- minbias events

- low momentum isolated tracks (10-15-25 GeV)

Both choices have + and - features

 symmetry with minbias events

The estimation of mean energy per readout:

The variance estimation:

The mean energy and variance in the ring:

<Ering> = <Esignal>+<Enoise>

<Vring>=<Vsignal>+<Vnoise>

Noise mean values and variances have to be estimated either from the same event using first time slices (startup scenario) or from pedestal runs (when pileup starts).

 symmetry with minbias events: calibration procedure

Coefficients are derived from non-calibrated sample for each

readout i in each -ring j after Noise subtraction

Mean values:

Cij = <Eij>/<Ej>

or variances:

Cij = sqrt(<Vij>/<Vj>)

Coefficients are applied and the new readout response is:

Note: to avoid noise/signal

correlations special runs

without zero-suppression

in HCAL

Eij corrected = Eij/Cij

Project was started in 2006 and was tested

with CSA06/CSA07/CSA08/CRAFT

 symmetry with minbias events: noise contribution

Ring:

Eta=1

depth=1

10 mln (noise+signal)

Signal is much

less then noise

in barrel

-index

2 mln (noise)

CSA07 data

-index

 symmetry with minbias events:

needed statistics estimation

Depends on the noise value, i.e. Most critical for the central barrel

ieta-ring<5.

1. Calibration with mean value: Mean noise = 10-5 with RMS=0.2-0.3 GeV

Mean signal

HB: 0.002 GeV (ieta=1)

HB: 0.008 GeV (ieta=14)

HE: 0.03 GeV (ieta=21)

HF: 0.5 GeV (ieta=35)

2. Calibration with variance: Mean Noise variance =0.084, RMS=0.24 GeV2

fluctuates channel-to-channel

Mean variance

HB: 0.005 GeV2 (ieta=1)

HB: 0.008 GeV2 (ieta=14)

HE: 0.059 GeV2 (ieta=21)

Error of noise RMS/sqrt(N)

< 0.02 * Esignal

N=25 mlns events for HB, ieta=1

Error of noise RMS/sqrt(N)

< 0.02 * Vsignal

N=6 mlns events for HB, ieta=1

Coefficients vs number of events in HF, ieta=32 (2-dim vs if)

900k

90k

8.9 mlns

1.8mlns

Coefficients (ideal calibration) with variances

for 9 mlns events

HB, ieta=1

HE, ieta=23

Material effects

Channels with large noise RMS

require higher statistics

(optimistic) or can not be included

in f-symmetry if noise distribution

is essentially non-gausian

HF, ieta=32

Accuracy vs number of events (IDEAL calibration)

h<0

Black: 90K

Red: 900K

Green: 1.8mln

Pink: 8.9mlns

Large noise channels

were extracted from the

calculations.

HE

HF

HB

The best accuracy

that can be achieved

with 9 mlns of events

and current noise map

Azimuthal symmetry jobs with CRAFT

L1 trigger : EG and Muon

AlCaRAW is created at HLT step to be sent to Tier0

only HCAL RAW data (without zero-suppression)

Special reconstruction at Tier0 followed by

AlCARECO production

Time slices 1-4 for noise reconstruction

Time slices 5-8 for signal reconstruction

160 mlns events are at CAF: same amount for noise and signal events.

50 mlns were analyzed:

4 days of data collected from 29st of Oct to 1st of Nov.

~2 days with HF on.

It can be used to study

the level of noise and

noise stability

Variance distributions in HB (50 mlns NZSP events)

Ieta=-1

iphi=70

HB

iphi

2

3

4

ieta

Var(ieta,iphi) with

selection:

Var=[0.22-0.26]

Same noise

map/gains

were in

CMSSW:

Slides 1-8

Iphi=70

ieta=-1

iphi=70, ieta=-1

due to small conversion factor.

Peak 4

E, GeV

Peaks origin (example with HB)

Peak 1

Peak 2

Peak 3

Peak 4

Problematic HPD, iphi=70

Var>0.1

Variance stability

From 29th Oct to 1st of Nov

HF was on later (on 30th Oct?)

HB, iph=70. ieta=-1

HF, ieta=36, iphi=3

Something

happened

HF, ieta=36

Variances are stable, but

sometimes we observe fluctuations

Phi symmetry with Isotracks

Method:

cell-by-cell calibration with isotracks

Problems:

- statistics for tracks with P>15 GeV/c

we need a few hundreds MIP tracks per cell

- material effects which depend on track energy

- shower profile vs energy ?

We will use the interval from 15 to 25 GeV or from

10 to 25 GeV

- Zero suppression affects low momentum tracks ?

Cell by cell calibration with 50 GeV

~45 tracks/cell

~22 tracks/cell

Material effects

~100 tracks/cell

From G.Safronov,

A.Anastassov

~900 tr/cell

Summary

Phi symmetry in high -h region (outside tracker) can be done with

minbias events with accuracy less than 1 %. For HB we can reach

3-3.5% in rings ieta<5 and 2% in ieta=5-14 and in HE. ~10 mlns events

have to be collected in special NZSP run.

The possibility was checked with CSA07/CSA08/Latest generation with

CMSSW219(IDEAL calibration)/CRAFT.

Phi symmetry in HB/HE can be done/cross-checked with isolated tracks

sample. However this possibility needs some additional study.

More study need to be done with Cosmic (noise stability) and with new

simulation with the latest noise/gains.

Latest presentations:

http://indico.cern.ch/materialDisplay.py?contribId=2&materialId=slides&confId=51006

http://indico.cern.ch/getFile.py/access?contribId=68&sessionId=2&resId=0&materialId=slides&confId=46352