Statistical Modeling of SiPM Noise - GSI Indico …...S. Vinogradov, IEEE NSS/MIC 2009, TNS 2011,...
Transcript of Statistical Modeling of SiPM Noise - GSI Indico …...S. Vinogradov, IEEE NSS/MIC 2009, TNS 2011,...
Statistical Modeling of SiPM Noise
Sergey Vinogradov
Lebedev Physical Institute of the Russian Academy of Sciences, Moscow, Russia
National Research Nuclear University «MEPhI», Moscow, Russia
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 1
Scope & outline
◙ Photon detection – a series of stochastic processes described by statistics
◙ SiPM response – a result of the stochastic processes – a random variable
◙ SiPM “noise” has a double meaning: as well as a “signal”
◙ Physics point of view – specific nuisance contributions to the response
◙ Dark counts - DCR
◙ Crosstalk - CT
◙ Afterpulsing – AP
◙ Statistics point of view – standard deviation of the response
◙ All above +…
◙ Multiplication – Gain
◙ Photo-conversion!!! – PDE
◙ Statistics of the response times
◙ Statistics of the response quantities
◙ Statistics of the response transients
◙
◙ Excess noise factor – ENF – as a Figure of Merit for all noise contributions
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 2
Definitions of statistics
◙ Statistics – wide sense – math related to random variables
◙ Random variable is fully described by its probability distribution Pr(X)
◙ Statistic – narrow sense – any function on probability distribution
Statistical moments nth order mn of random variable X:
― 𝑚𝑛(𝑋) = σ𝑖=0∞ 𝑖𝑛 Pr(𝑋 = 𝑖) X – discrete r.v.
― 𝑚𝑛(𝑋) = ∞−∞𝑥𝑛 Pr 𝑋 = 𝑥 𝑑𝑥 X – continuous r.v.
― first moment Mean (µ) = 𝑚1 ;
― second central moment – Variance (σ2) = 𝑚2 − 𝑚12
μ
σ
𝑷𝒓 𝑿 = 𝒙
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 3
Statistics of random times
◙ How probability distribution of signal and noise event times
are transformed by photon detection processes
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 4
Statistics of random times
https://doi.org/10.1016/j.nima.2015.07.009
Pr(∆t) or PDF– full characterization
Mean and Variance – not so obvious
because affected by recovery losses
and correlated events
~ Poisson affected by dead time
~ Poisson affected by afterpulsing
Pure Poisson – exponential distribution of ∆t
Poisson process:
No events in : Pr( | 0)
Interval between events : Pr( | 0) 1
Probability density function: ( )
Mean time 1/ Standard deviation ( ) 1/
DCR t
DCR t
DCR t
t t e
t e
PDF t DCR e
t DCR t t DCR
−
−
−
=
= −
=
= = =
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 5
Analysis based on probability density function (PDF)
= histogram of interarrival times
◙ Common-sense analysis of DCR and correlated events
◙ Contributions are fitted separately in specific time frames
P. Eckert et al, NIMA 2010
F. Acerbi et al, TNS 2016
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 6
Analysis based on cumulative distribution function (CDF)
= list of interarrival times
◙ Complimentary CDF approach: probabilities of independent “zero”
dark and “zero” correlated events in ∆t are multiplied:
( ) ( )
( )
.......
safterpulse andcrosstalk of separationfor
approach (C)CDF extendfurther can weMoreover,
shape itson sassumptionany without ),(or ),( analyse And
)exp(),,(11),(
: ),( events correlated ofon distributi pure
extract toallgot ve we'now So,
)exp(1)(),(1),(1),,(1
CDFary compliment1
APCTcorr
corrcorrcorrcorr
phtotalcorrcorr
corrcorr
dark
darkcorrcorrcorrtotal
darkcorrtotal
CCDFCCDFCCDF
PtfPtF
tDCRDCRNtFPtF
PtF
tDCRtFDCRtFPtFDCRPtF
CDFCCDFCCDFCCDFCCDF
=
−−=
−−=−−=−
−==
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 7
CCDF time distribution:
dark and correlated events
Pco
rr1
-P
co
rr
Experimental example – courtesy of E. Popova, D. Philippov… , NSS/MIC 2016
Correction on lost
dark counts at Tmax
should be applied:
)exp( max_ TDCRCCDF correctiondark −=
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 8
CCDF time distribution:
reconstruction of correlated event CDF
Pco
rr
Now we are ready to play with correlated
event CDF or PDF applying any models
and analysis.
For example, typical assumption on
exponential time distribution, but take care
about multiple exponents (see backup):
moreandeP
Ptf
moreandePPtF
corr
corr
t
corr
corrcorrcorrcorr
t
corrcorrcorrcorr
...),,(
...1),,(
−
−
=
−=
( ) )exp(),,(11),( tDCRDCRNtFPtF phtotalcorrcorr −−=
Experimental example – courtesy of E. Popova, D. Philippov , NSS/MIC 2016
More details – next talk by Eugen Engelmann @ ICASiPM
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 9
Empirical CDF (list mode) vs PDF (histogram):highest precision due to full information from all data points
◙
[1] Y. Cao, L. Petzold, “Accuracy limitations and the measurement of errors in the stochastic simulation…”, J. Comput. Phys. 212, 2006
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 10
◙ How probability distributions of signal and noise quantities
are transformed from input to output
Statistics of random quantities
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 11
◙ Pr(Nout|Nin) – full characterization
◙ Mean and Variance – partial characterization
Statistics of random quantities
Input photon distribution Photodetection processes Output charge distribution
Pr(Nin)
Pr(Nout)
Process i
Process j
)( inN
inN outN
)( outN
Photons (Poisson) photoconversion (Bernulli) photoelectrons (Poisson)
Dark+photoelectrons (Poisson) triggering (Bernulli) primary avalanches (Poisson)
Primary avalanches (Poisson) , (TBD) secondarCT AP
y avalanches (TBD)
All avalanches(TBD) multiplication (TBD) output electrons (TBD)
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 12
Correlated stochastic processes of CT & AP
Branching
Poisson
Process
Geometric
Chain Process
Poisson number of primaries <N>=μ
e.g. SiPM photoelectron spectrum
Single primary event N≡1
e.g. SiPM dark electron spectrum
Process
Models
Non-random (Dark) event
Primary
1st event
2nd event
No event
Ran
dom
AP
even
ts…
Ra
nd
om
CT
ev
ent
s
Random Photo eventsRandom primary (Photo) events
Ran
dom
AP
even
ts
…
…
…
Non-random (Dark) event
Ran
dom
CT
even
ts
…
… …
…
…
…
Random primary (Photo) events
Ran
dom
CT
even
ts
…
…
λµ
S. Vinogradov, IEEE NSS/MIC 2009, TNS 2011, NDIP 2011
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 13
Models Geometric chain process Branching Poisson process
Primary event
distribution
Non-random
single (N≡1)
Poisson (μ) Non-random single
(N≡1)
Poisson (μ)
Total event
distribution Geometric (p)
Compound
Poisson (μ, p) Borel (λ) Generalized Poisson (μ, λ)
P(X=k) ( )ppk −− 11
−
−
sp
s
eL 1
1~
( ) ( )!
exp1
k
kkk
−−
( ) ( )
!
exp1
k
kkk
−−+−
E[X] p−1
1
p−1
−1
1
−1
Var[X] 2)1( p
p
−
2)1(
)1(
p
p
−
+
( )31
−
( )31
−
ENF p+1 )...(2
31
)1ln(1
1
1
1 32 pppp
+++−+
=−
1.0E+01
1.0E+02
1.0E+03
1.0E+04
1.0E+05
1.0E+06
1.0E+07
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Threshould [fired pixels]
Dark
Co
un
t R
ate
[H
z]
Experiment
Borel
Geometric
Analytical results for CT & AP statistics
Pct=40%
Pct=10%
S. Vinogradov, NDIP 2011 (experiment – R. Mirzoyan, 2008 (MEPhI SiPM)) S. Vinogradov, NDIP 2011 (experiment – LPI, Hamamatsu MPPC)
0%
10%
20%
30%
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
k, number of events
Pro
bab
ilit
y o
f k e
ven
ts
p=0 L=3
p=0.2 L=3
p=0.5 L=3
Crosstalk
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 14
Statistics of transient signals - stochastic functions
◙ How random quantities are developed in time
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 15
Statistics of filtered marked correlated point processes
Signal IntensityDark noise,
Background
Dark eventCorrelated eventDark event
Single Electron Response (IRF)
Random Gain (amplitude)
False signal detectionDiscriminator
True signal detection
Signal events
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 16
Mean and Variance of
filtered marked Poisson point process
dt
thdN
V
VthENFN
dt
tVoutd
tVVar
VthENFVtVVar
tPDENtRGainq
VthVtVoutE
specificSIPM
tXVartttXCOVREMARK
thA
AttIRFCOVXCOVtYVar
tdtNthAtIRFEtXEtYE
NiiidrvAthAtIRFttIRFtY
NiiidrvtPoissonianNtttX
sersptrph
pe
ser
noisesersptrphgainpe
out
t
noisegainserout
sptrphph
fall
loadserser
ttinin
A
ttinout
t
inout
ii
N
i
iout
i
N
i
iin
)(
)(
)(
)]([
)()]([
)()()()]([
)]([)()(][:
)(1)]([][)]([
)()()]([)]([)]([
)...1()()()()(
)...1()()(
2
22
222
2
2
22
0
1
1
+
==
+=
=
=
→−=
+=−=
===
=−=−=
=−−−=
→
=
=
=
/ /
2
Example: baseline fluctuation due to DCR
( ) ( )
Var =V2
fall riset t
fall riseserout
t DCR const h t e e
V ENF DCR
− −= = = −
+
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 17
Figure of Merit for all noise contributions
◙ How to compare noises and optimize SiPM operations:
Excess Noise Factor – ENF – approach
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 18
ENF: two definitions
◙ ENF in multiplying photodetectors (APD, PMT):
◙ ENF in amplifying electronics:
Noise factor (F) measures degradation of the signal-to-noise ratio (SNR), caused by components in a signal chain.
2
2
22
1
( )1
( )
Mm MENF
m M M
= = +
2 22 2
2 2
in outout in
in out
ENF Res Res
= =
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 19
ENF: two definitions are equal for Poisson input
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 20
2 2
2 2
1( 1) 1 ( ) 1
( )
Input quantity N Poisson ( ) ( 1)
M Min in
in in
ENF X vs ENF X NFano NM M
ENF X N ENF X
= + = = +
= =
Resolution at output is a product of Resolution at input and √ENF(Xin≡1)
Total ENF for a sequence of specific processes is approximately
ENF as a measure of SNR (resolution) degradation
)1()(
1)()(
1
)1(
)1(1)()(
2
2
=
=
+= inin
in
inout
outinout XENFXRES
XFanoXFanoY
XRESYRES
...)()()( 21 processprocessintotalinout ENFENFXRESENFXRESYRES ==
Input photon distribution Photodetection processes Output charge distribution
RES(Xin)
RES(Yout)= RES(Xin)√ENF
ENFi
ENFj
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 21
ENF related to Gain, PDE and DCR
◙ Multiplication (single electron)
◙ Photon detection (single photon)
◙ Dark counts (Poisson dark+pe)
pedcr N
tDCRENF += 1
PDEPDE
PDEPDEENFpde
1)1(1
2=
−+=
SiPM ofmost for 05.1...01.1
12
2
+=
gain
gain
gain
ENF
GainENF
Xin=1
PDF(Yout) ?
PDF(Xin)=δ(x)
<Yout>=Gain
Photoelectrons
RES(Yout)
Dark counts
Xin=1
PDF(Yout) = BernoulliPDF(Xin)=δ(x)
<Yout>=PDE
<Yout>=Npe+Nd
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 22
Pr(0)
Methods of ENF measurements
1
10
100
1000
10000
100000
1000000
10000000
0 1 2 3 4 5 6 7 8 9 10
Amplitude [Single Electron Response]
Eve
nts
Crosstalk amplitude histogram
N=1
N=1 non-random primary
_________________________________
N ϵ Poisson random primaries
Nph could also be measured by reference PD
22 2
2 2 2( )
ln(Pr(0))
outin outin
in out out
ENF X N N
N
= = =
= −
2
2( 1) 1 out
in
out
ENF X
= +
Remark: ENF measurements in electronics: noise generator + noise power meter
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 23
σ/µ
σ/µ
Key SiPM parameters in ENF metrics
Noise source Key parameter Typical values for
SiPM ENF expression
SiPM
ENF
PMT
ENF
Fluctuation of
multiplication
Mean gain and standard
deviation of gain
μ(gain) ~ 106
σ(gain) ~ 105 )(
)(1
2
2
gain
gainFm
+= 1.01 1.2
Crosstalk Probability of crosstalk
event Pct 5% – 40%
1
1 ln(1 )Fct
Pct=
+ − 1.05 – 2 1
Afterpulsing Probability of afterpulse
Pap 5% – 20% 1Fap Pap= + 1.05 – 1.2 1.01
Shot noise of dark
counts
Dark count rate DCR
Signal events in Tpulse 105 – 106 cps 1
DCR TpulseFdcr
Nph PDE
= +
1.01 1
Noise / losses of
photon detections
Photon detection
efficiency PDE 20% – 50%
1Fpde
PDE= 2 – 5 3 – 5
Total noise in linear
dynamic range Total ENF_linear _ENF linear Fm Fct Fap Fdcr Fpde= 3 – 12 4 – 6
Binomial
nonlinearity
Mean number of possible
single pixel firings n
exp( ) 1; ; binomial distribution
Nph PDE nn ENFpix
Npix n
−= =
Dead time
nonlinearity
Mean rate of possible
single pixel firings λ ; 1 ; nonparalizible model
nENFrec Trec
Tpulse = = +
total
nlapctdcrpdemtotaltotal
ph
ph
out
out
ENFDQEFFFFFFENFENF
N
N
N
NPNR
1
)(
)(
)(
)(
*
* ===
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 24
The end
Thank you for your attention!
Questions?
Objections?
Opinions?
…
Sergey Vinogradov Statistical Modeling of SiPM Noise ICASiPM 11-15 June 2018 Schwetzingen, Germany 25
FBK results on Generalized Poisson model
A. Gola, A. Ferri, A. Tarolli, N. Zorzi, and C. Piemonte, “SiPM optical crosstalk amplification due to scintillator crystal: effects on timing
performance,” Phys. Med. Biol., vol. 59, no. 13, p. 3615, 2014.
Sergey Vinogradov LIGHT – 2017 19 October 2017 26
DESY results on Generalized Poisson
◙ E. Garutti group @ DESY: V. Chmill et al., NIMA 2017
Sergey Vinogradov LIGHT – 2017 19 October 2017 27