22
Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Embed Size (px)

Citation preview

Page 1: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Noise Analysis in NaIAD

Matthew Robinson

Experiment now running reliably therefore time to work on getting the best from the analysis

Experiment now running reliably therefore time to work on getting the best from the analysis

Page 2: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Noise ExperimentMeasurements taken with and without light guides in place.

Taken by Phil and Mike in glove box @ sheffield

Measurements taken with and without light guides in place.

Taken by Phil and Mike in glove box @ sheffield

Page 3: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Anomalous Events

Looks Like Scintillation but isn’t.

Many cut by asymmetry but also many left behind

Looks Like Scintillation but isn’t.

Many cut by asymmetry but also many left behind

Page 4: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Time Constant Distribution

Distribution with light guides

@ 4-6 keV

Energy calibration not possible, typical values used:

20 mV/keV (sum)

Poor fit to power law as shown,

Terrible fit to exponential,

Work on cuts to identify scintillation like noise events in data

Distribution with light guides

@ 4-6 keV

Energy calibration not possible, typical values used:

20 mV/keV (sum)

Poor fit to power law as shown,

Terrible fit to exponential,

Work on cuts to identify scintillation like noise events in data

Page 5: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Time Constant Distribution

Distribution without light guides.

Appears similar, now collecting more data for comparison

Distribution without light guides.

Appears similar, now collecting more data for comparison

Page 6: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Identification of Noise EventsMeasurement

Measurement Region

Now measured from fitted start of event to end of data rather than from 0 ns

Now measured from fitted start of event to end of data rather than from 0 ns

Page 7: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Identification of Noise EventsMeasurement

2

2

)(

end

tt datastart

datafit

data 1d.o.f

Set event by event to achieve

Uncertainty for each fit proportional to amplitude of event. This achieves a reasonable per degree of freedom value for chi squared

Uncertainty for each fit proportional to amplitude of event. This achieves a reasonable per degree of freedom value for chi squared

Page 8: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Identification of Noise Events

short Measurement

Measurement Region

Measured similarly to normal chi squared except over first 1000 ns of event.

Flat part of event contains few photons and is therefore less important

Measured similarly to normal chi squared except over first 1000 ns of event.

Flat part of event contains few photons and is therefore less important

Page 9: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Identification of Noise Events

short Measurement

21000

2

)(

nst

tt data

start

start

datafit

data As for

Page 10: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Identification of Noise EventsSteppiness Measurement

Measurement Region

Measurement of tendency to be made up of large steps.

Measurement of tendency to be made up of large steps.

Page 11: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Identification of Noise EventsSteppiness Measurement

N

n

n

pen pe

pe2

sumin termsofnumber

ronsphotoelect ofnumber expected

ronsphotoelect ofnumber

N

n

n

pe

pe

Expected number of photoelectrons calculated using differential of fit at that point.

Steps found by searching through event one time bin at a time and counting the number of photons in the next 30 ns.

Expected number of photoelectrons calculated using differential of fit at that point.

Steps found by searching through event one time bin at a time and counting the number of photons in the next 30 ns.

Page 12: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Identification of Noise EventsAfter-pulses and Jitter

Measurement Region

Jitter originally included to measure digitisation noise, now used to spot after-pulses which tend to occur in this region.

Jitter originally included to measure digitisation noise, now used to spot after-pulses which tend to occur in this region.

Page 13: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Identification of Noise EventsJitter Measurement

2

10002

)(

end

nstt datastart

datafit

data As for

Basically a chi squared measured over what should be the flat part of the data.

If the value is significant, an attempt can be made to fit just the early part of the data.

Basically a chi squared measured over what should be the flat part of the data.

If the value is significant, an attempt can be made to fit just the early part of the data.

Page 14: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Identification of Noise EventsLimited Time Fit

Alternatively, events with after-pulses may be cut completely by cutting on the jitter parameter

Alternatively, events with after-pulses may be cut completely by cutting on the jitter parameter

Page 15: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Comparison of Cuts

Uncut Noise

Far too many events with time const >100 ns

Far too many events with time const >100 ns

Page 16: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Comparison of Cuts

Noise Cut by Asymmetry

Long time constant events reduced but not eliminated

Asymmetry in time const. < 100 ns,

Asymmetry in start time < 100 ns.

Asymmetry in energy < 40%

Long time constant events reduced but not eliminated

Asymmetry in time const. < 100 ns,

Asymmetry in start time < 100 ns.

Asymmetry in energy < 40%

Page 17: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Comparison of CutsNoise Cut by

short

Long time constant events reduced but not eliminated

Asymmetry in time const. < 100 ns,

Asymmetry in start time < 100 ns.

Asymmetry in energy < 40%

Long time constant events reduced but not eliminated

Asymmetry in time const. < 100 ns,

Asymmetry in start time < 100 ns.

Asymmetry in energy < 40%

Page 18: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Comparison of Cuts

Noise Cut by Steppiness

Long time constant events reduced but not eliminated

Asymmetry in time const. < 100 ns,

Asymmetry in start time < 100 ns.

Asymmetry in energy < 40%

Long time constant events reduced but not eliminated

Asymmetry in time const. < 100 ns,

Asymmetry in start time < 100 ns.

Asymmetry in energy < 40%

Page 19: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Comparison of Cuts

Noise Cut by Identification

Cuts optimised to achieve same photopeak size in data (later slide)

Chi squared/d.o.f <3 all channels,

Short chi squared/dof <2 (sum) <2.5 channels,

Steppiness<4 (sum) <5 (channels)

Much better reduction of noise

Cuts optimised to achieve same photopeak size in data (later slide)

Chi squared/d.o.f <3 all channels,

Short chi squared/dof <2 (sum) <2.5 channels,

Steppiness<4 (sum) <5 (channels)

Much better reduction of noise

Page 20: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Comparison of Cuts

Uncut Data

Not really possible to use data in this condition.

Not really possible to use data in this condition.

Page 21: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Comparison of Cuts

Data Cut by Asymmetry

Asymmetry cuts have greatest effect on fast noise, which we don’t care about.

Asymmetry in time const. < 100 ns,

Asymmetry in start time < 100 ns.

Asymmetry in energy < 40%

Asymmetry cuts have greatest effect on fast noise, which we don’t care about.

Asymmetry in time const. < 100 ns,

Asymmetry in start time < 100 ns.

Asymmetry in energy < 40%

Page 22: Noise Analysis in NaIAD Matthew Robinson Experiment now running reliably therefore time to work on getting the best from the analysis

Comparison of Cuts

Data Cut by Identification

Fit to asymmetry cut tcd shown here for comparison.

Identification cuts have little effect on fast noise, which make it easier to use in fitting of noise.

Chi squared cuts tend to increase mean time constant, steppiness cuts have opposite effect. This can be used to persuade data and compton distributions to match.

Probably decide to use both identification and asymmetry, have to do more work on optimising cuts.

Chi squared/d.o.f <3 all channels,

Short chi squared/dof <2 (sum) <2.5 channels,

Steppiness<4 (sum) <5 (channels)

Much better reduction of noise

Fit to asymmetry cut tcd shown here for comparison.

Identification cuts have little effect on fast noise, which make it easier to use in fitting of noise.

Chi squared cuts tend to increase mean time constant, steppiness cuts have opposite effect. This can be used to persuade data and compton distributions to match.

Probably decide to use both identification and asymmetry, have to do more work on optimising cuts.

Chi squared/d.o.f <3 all channels,

Short chi squared/dof <2 (sum) <2.5 channels,

Steppiness<4 (sum) <5 (channels)

Much better reduction of noise