This document discusses sensor modeling and photometry techniques applied to astrophotography. It presents a case study imaging the Triangulum Galaxy using a modified DSLR camera and telescope. Key steps included estimating signals and noises from light, dark, and offset frames to characterize the sensor and extract measurements of sky background, dark current, and target object luminance. Expectations and variances of photoelectrons generated were related to measured analog-to-digital units. The techniques allowed optimizing exposure times based on read noise, guiding error, and dynamic range criteria.
Sensor modeling and Photometry: an application to Astrophotography
1. Sensor modeling and Photometry:
an application to Astrophotography
Laurent Devineau
2. Introduction
⪠The objective is to present with a case study, key notions of photometry and sensor modeling
applied to astrophotography.
⪠After an introduction to different signal and noise categories captured by a camera, we deal with
time exposure optimization of an astrophotography session .
⪠We detail methodologies based on Read Noise dominance, on guiding error and on loss of dynamics
management. This allows to get optimal time exposure regarding these different criteria.
⪠We also present signal to noise ratio (SNR) calculation techniques of a stacked image and we
formalize a methodology to determine global exposure time to reach a target SNR.
⪠Some theoretical developments are also proposed to justify the Poissonian nature of the photonic
signal and to prove the expectation / variance equality of the photoelectrons generated by the
sensor.
2
3. Case study description
⪠The techniques detailed in this document are illustrated with a
case study.
⪠This case study consisted in imaging the Triangulum Galaxy (M33)
with the following framework:
⪠Digital Single Lens Reflex (DSLR) Canon 500 D modified with a sensor
CMOS and a pixel size of 4.68 µm ;
⪠Newtonian telescope(*) of 1000 mm focal length and a primary mirror of
200 mm (ð¹/ð· = 5 aperture).
⪠Below the details of image series:
⪠80 unitary light frames of 240 seconds at ISO 400, for a global exposure
time of 5h20â
⪠A series of DOFs (Darks Offsets Flats) consisting of :
⢠45 dark images of 240 seconds at ISO 400
⢠31 flat images of 1/3 second with a ISO 100 gain (in order to
concentrate the histogram on 2/3)
⢠101 offset images of 1/4000 second at ISO 400
3
Image pretreated after integration
Example of a light image
(*) In order to reduce light pollution, an Optolong L-Pro filter has been integrated into the optical path.
5. Reminder of digital sensor specificities (1/2)
5
Photons
Photons strike the
photosite
Some electrons are generated,
they fall into the well
1
2
Measure
At the end of the exposure
time, the electrical voltage is
measured and the ADC
converts it into a digital value
(ADU)
3
ADC(*)
(*) Analog / digital converter
214 â 1
0
11 000
11 000 ADU Display
The numerical values ââof each
photosite are saved in a file
4
⪠Photoelectron generation process and digital encoding:
6. Reminder of digital sensor specificities (2/2)
⪠The term ADU (Analog to Digital Units) corresponds to the digital value
produced by the ADC (Analog to Digital Converter) from the electrical
voltage of the device.
⪠The ADUs are encrypted on a scale which depends on the dynamic range of
the sensor expressed in bits. In the case study, the sensor encodes in 14
bits, the ADUs therefore take (integer) values ââbetween 0 and 214 â 1.
⪠The ADU value is deduced from the number of electrons generated from a
gain parameter denoted by ððŽð·ð/ðâ. Conversely, when we know the ADU
value of the pixel, we estimate the number of electrons generated with a
gain parameter denoted by ððâ/ðŽð·ð. We have: ððŽð·ð/ðâ = 1/ððâ/ðŽð·ð
⪠The electron capacity (full well capacity) represents the maximum number
of electrons that can be stored in the well. The smaller this quantity, the
faster the image saturates.
⪠The Quantum Efficiency is the ratio between the number of electrons
generated and the quantity of photons reaching the sensor.
6
Illustration
Gain shape of EOS 500D
7. Signals and noises
⪠The signals recorded in the image capture process are not deterministic but variables (presence of
noise around the expected value).
⪠There are different types of signals (measured in ADU):
⢠The signal ðððð
ðŽð·ð
(ð¡) of target object,
⢠The signal ðð¿ð
ðŽð·ð
(ð¡) related to light pollution (sky background luminosity),
⢠The Dark Current (DC) signal ðð·ð¶
ðŽð·ð
(ð¡) of the sensor,
⢠The bias signal (offset) ðð ð
ðŽð·ð
related to the presence of an offset between the pure black level
reference and the non-zero value returned by the sensor. The noise component associated to this
signal is called Read Noise (RN).
⪠Denoting ðð¡ðð¡
ðŽð·ð
ð¡ the total signal, we therefore have: ðð¡ðð¡
ðŽð·ð
ð¡ = ðððð
ðŽð·ð
(ð¡) + ðð¿ð
ðŽð·ð
(ð¡) + ðð·ð¶
ðŽð·ð
(ð¡) + ðð ð
ðŽð·ð
⪠Under the same notations for the standard deviations, and assuming the independence of the
random sources, it comes:
ðð¡ðð¡
ðŽð·ð
= ðððð
ðŽð·ð
(ð¡)2 + ðð¿ð
ðŽð·ð
(ð¡)2 + ðð·ð¶
ðŽð·ð
(ð¡)2 + ðð ð
ðŽð·ð
²
7
8. Link between electrons numbers and ADU measures
⪠The signals introduced previously can also be measured in number of electrons (e-). For example consider
the signal (in e-) of the target object, denoted by ðððð
ðâ
ð¡ . We therefore have the relation:
ðððð
ðâ
ð¡ = ððâ/ðŽð·ð à ðððð
ðŽð·ð
ð¡
⪠It is possible to prove under relevant assumptions (cf. appendix) that for the signals ððð and ð¿ð, the
expectation and the variance of the number of electrons generated are equal and linear in ð, the exposure
time.
⪠We write: ðž ðð
ðâ
ð¡ = ð ðð
ðâ
ð¡ = ðð
ðâ
. ð¡ and ð = ððð, ð¿ð
⪠From previous equation, it is straightforward to deduce that: ððâ/ðŽð·ð =
ðž ðð
ðŽð·ð
ð¡
ð ðð
ðŽð·ð ð¡
⪠Under the same assumptions we prove that the terms ð¬ ðºð¿
ðšð«ðŒ
ð ððð ðœ ðºð¿
ðšð«ðŒ
ð are linear in ð. We have:
ðž ðð
ðŽð·ð
ð¡ =
ðð
ðâ
ððâ/ðŽð·ð
à ð¡ ððð ð ðð
ðŽð·ð
ð¡ =
ðð
ðâ
ððâ/ðŽð·ð
2
à ð¡
Remark: variance ð ðð·ð¶
ðŽð·ð
ð¡ is also assumed to be linear in ð¡.
8
Instantaneous rate
related to signal ð
9. Estimation on light, dark and offset images (1/6)
⪠The estimation of different signals expectations and standard
deviations was performed by the process Statistics of PixInsight
software.
⪠For the offset signal, the standard deviation was evaluated on 2
images:
ðð ð
ðŽð·ð
=
1
2
ð ðŒðððð ðð¡
1
â ðŒðððð ðð¡
2
+ ð¶
⢠Where ð¶ is a translation factor to avoid truncating negative data
⪠Estimation of offset signal expectation and standard deviation:
9
Offset signal Channel R Channel G Channel B
ðð ð
ðŽð·ð 13.7 9.9 11.5
ðð ð
ðŽð·ð 10.5 6.6 8.3
10. Estimation on light, dark and offset images (2/6)
⪠The dark signal implicitly integrates the offset signal. To estimate its expectation and standard
deviation, the following adjustments are made:
ðð·ð¶
ðŽð·ð
=
1
2
ð ðŒðððð
1
â ðŒðððð
2
+ ð· â ðð ð
ðŽð·ð2
Et
ðð·ð¶
ðŽð·ð
= ð ðŒðððð
ð
â ðð ð
ðŽð·ð
, ð = 1 ðð¢ 2
⪠Estimation of dark signal expectation and standard deviation:
10
Dark signal Channel R Channel G Channel B
ðð·ð¶
ðŽð·ð 18.6 13 15.2
ðð·ð¶
ðŽð·ð 10.7 8.3 9.7
11. Estimation on light, dark and offset images (3/6)
⪠We also want to perform an estimation of the average signal ðð¿ð
ðŽð·ð
and the sky background (SB) noise ðð¿ð
ðŽð·ð
⪠The SB signal integrates the dark and offset signals. To estimate its
expectation and standard deviation, the following adjustments
are made:
ðð¿ð
ðŽð·ð
= ððŽ
ðŽð·ð2
â ðð·ð¶
ðŽð·ð2
â ðð ð
ðŽð·ð2
And
ðð¿ð
ðŽð·ð
= ð ðŒðððâð¡ÈðŽ
â ðð·ð¶
ðŽð·ð
â ðð ð
ðŽð·ð
⪠The term ð ðŒðððâð¡ÈðŽ
corresponds to the mean value of the signal
measured on an area of ââSB in the light frame considered. This is a
measure of light pollution (LP):
11
SB signal Channel R Channel G Channel B
ðð¿ð
ðŽð·ð
1694.6 1706.3 1763.6
ðð¿ð
ðŽð·ð 67.9 57.5 65.4
12. Estimation on light, dark and offset images (4/6)
⪠It is possible to estimate the signal of the target object. We
favor a peak luminosity rather than an average luminosity
because it constitutes a more relevant criterion for the
evaluation of the signal / noise ratio.
⪠To estimate the peak luminosity, we select the brightest
area of ââthe object (i.e. the core of the galaxy) which we
reprocess by removing the signal from the central star
(denoted by ðâ
ðŽð·ð) as well as the other signals (light
pollution, RN and DC):
ðððð
ðŽð·ð
= ð ðŒðððâð¡Èððð
â ðâ
ðŽð·ð
â ðð·ð¶
ðŽð·ð
â ðð ð
ðŽð·ð
â ðð¿ð
ðŽð·ð
⪠Below the estimations of peak luminosities:
12
Signal of Object Channel R Channel G Channel B
ðððð
ðŽð·ð
342.9 354.5 379.2
13. Estimation on light, dark and offset images (5/6)
⪠For sources ð = ð¿ð ðð ððð, we can deduce the expectation of
number of photons ðð
ðâ
which reach the sensor, by using ADU
measurements carried out:
ðð
ðâ
=
1
ððž
Ã
ðð
ðŽð·ð
ð¡ððð
à ððâ/ðŽð·ð Ã
1
ð à ð ðµððŠðð
ð€ðð¡â ð = ð¿ð ðð ððð
⪠Under the notations: ððž = 38% the quantum efficiency, ð¡ððð =
240ð the unitary exposure time, ð = 73% the transmission rate of
the L-Pro filter, ð ðµððŠðð = 1/3 the transmission rate related to the
matrix of Bayer, ððâ/ðŽð·ð = 0.44 the gain in e-/ADU
⪠The average numbers of photons per second and per photosite are
as follows:
13
Photon
Probability of generating a
photoelectron â¡ quantum efficiency
ðžð¬ = ðð%
Number of
photons
Channel R Channel G Channel B
ðð¿ð
ðâ
34.0 34.3 35.4
ðððð
ðâ
6.9 7.1 7.6
14. Estimation on light, dark and offset images (6/6)
⪠In order to validate the Poissonian nature of the photonic signal of the Sky Background, we want to
verify the aforementioned equation:
ððâ/ðŽð·ð =
ðž ðð¿ð
ðŽð·ð
ð¡
ð ðð¿ð
ðŽð·ð
ð¡
⪠Under the notations and adjustments previously introduced, it consists in proving that the term ෡
ð®
below is equivalent to the gain ððâ/ðŽð·ð = 0.44:
à·
ðº =
ð ðŒððð¢ð¡ðÈðŽ
â ðð·ð¶
ðŽð·ð
â ðð ð
ðŽð·ð
ððŽ
ðŽð·ð2
â ðð·ð¶
ðŽð·ð2
â ðð ð
ðŽð·ð2
⪠The results obtained below confirm the Poissonian hypothesis:
14
Gain
parameter
Channel R Channel G Channel B
à·
ðº 0.37 0.52 0.41
Average 0.43
ððâ/ðŽð·ð 0.44
Result very close to
the target value
15. Signal to Noise Ratio and sub exposure time optimization (1/5)
⪠We seek to optimize the sub exposure time per frame according to criteria associated with the Signal
to Noise ratio (SNR)
⪠We propose to carry out the analysis in several stages:
⪠Reminders on the growth of the SNR according to the number of images;
⪠Sensitivity of the SNR to the sub exposure time;
⪠Optimization of the sub exposure time by Read Noise dominance criterion.
⪠To develop these different notions we will consider simplified SNRs that we will enrich in the rest of
the presentation.
15
16. Signal to Noise Ratio and sub exposure time optimization (2/5)
⪠We recall here the results on the growth of the SNR as a function of the number of images considered.
⪠Below the SNR of a unitary image composed only of the signal of the object and the Read Noise:
ððð 1(ð¡) =
ðððð
ðâ
. ð¡
ðððð
ðâ
. ð¡ + ðð ð
ðâ
²
Where ðððð
ðâ
corresponds to the instantaneous rate associated with the signal of the object
⪠Suppose we stack ðµ frames of exposure duration ð¡. So the signal is:
ðð =
1
ð
à·
ð=1
ð
ðððð
ðâ
ð + ðð ð
ðâ
(ð)
⪠We get:
ððð ð ð. ð¡ =
ðž ðð
ð ðð
=
ðððð
ðâ
. ð¡
1
ð ðððð
ðâ
. ð¡ + ðð ð
ðâ
²
= ð Ã ððð 1(ð¡)
16
The SNR increases by a factor ðµ compared to the ðºðµð¹ð => we exhibit a result well known to
astrophotographers
Assumed centered by
removing the average offset
signal: E ðð ð
ðâ
(ð) = 0
17. Signal to Noise Ratio and sub exposure time optimization (3/5)
⪠Let us consider an overall exposure time budget of ð => this budget can be segmented in different ways.
⪠Consider two sub exposure durations ð¡1 and ð¡2 with ð¡1 < ð¡2. The associated unitary frame numbers are
denoted ð1 and ð2 and satisfy:
ð1. ð¡1 = ð2. ð¡2 = ð ððð ð2 < ð1
⪠Case where the read noise is zero:
ððð ð1
(ð) = ð1
ðððð
ðâ
. ð¡1
ðððð
ðâ
. ð¡1
= ðððð
ðâ
. ð1. ð¡1 = ðððð
ðâ
. ð = ðððð
ðâ
. ð2. ð¡2 = ððð ð2
(ð)
⪠General case:
ððð ð1
ð = ð1
ðððð
ðâ
. ð¡1
ðððð
ðâ
. ð¡1 + ðð ð
ðâ
²
=
ðððð
ðâ
. ð1ð¡1
ðððð
ðâ
. ð1ð¡1 + ð1. ðð ð
ðâ
²
=
ðððð
ðâ
. ð
ðððð
ðâ
. ð + ð1. ðð ð
ðâ
²
<
ðððð
ðâ
. ð
ðððð
ðâ
. ð + ð2. ðð ð
ðâ
²
= ððð ð2
(ð)
17
The segmentation strategy has no impact.
Only the total exposure time matters.
Read noise is counted as many times as sub frames are produced. In conclusion, the longest possible
exposures should be made (i.e. ð) even if in practice it is impossible for technical reasons related to the
realization of the frames (tracking errors, occurrences of planes or satellites,... ).
18. Signal to Noise Ratio and sub exposure time optimization (4/5)
⪠We seek here to optimize the sub exposure time by read noise dominance criterion(*). Consider the
expression of the SNR below in which we introduce the noise associated with light pollution:
ððð 1 ð¡ =
ðððð
ðâ
. ð¡
ðððð
ðâ
. ð¡ + ðð¿ð
ðâ
. ð¡ + ðð ð
ðâ
²
=
ðððð
ðâ
. ð¡
ðððð
ðâ
. ð¡ + ðð¿ð
ðâ
. ð¡. 1 +
ðð ð
ðâ
²
ðð¿ð
ðâ
. ð¡
⪠The objective is, given a level of tolerance ð¶, to find the minimum time, denoted ð¡ððð, such that for ð¡ â¥
ð¡ððð we have:
ðð¿ð
ðâ
. ð¡. 1 +
ðð ð
ðâ
²
ðð¿ð
ðâ
. ð¡
†1 + ðŒ ðð¿ð
ðâ
. ð¡
⪠We get:
ðððð =
ð
ð + ð¶ ð â ð
Ã
ðð¹ðµ
ðâ ð
ðð³ð·
ðâ
18
(*) Method implemented by Robin Glover the developer of the SharpCap astrophotography software
For ð¡ ⥠ð¡ððð the residual noise (i.e. in addition to the variance of the object signal) is at most
equal to the noise induced by the light pollution increased by the threshold ð¶.
19. Signal to Noise Ratio and sub exposure time optimization (5/5)
⪠To estimate ð¡ððð, we use the ADU measurements
detailed above with the equation:
ð¡ððð =
1
1 + ðŒ 2 â 1
Ã
ðð ð
ðŽð·ð2
ðð¿ð
ðŽð·ð
ð¡ððð
2
/ð¡ððð
Where ð¡ððð represents the reference time of the sub frame
used to estimate ðð¿ð
ðŽð·ð
(in the case study ð¡ððð =240sec).
⪠Below the values ââassociated with the 2% and 5%
thresholds:
19
Threshold ð¡ððð(ð ) ð¡ððð(ðº) ð¡ððð(ðµ) ððð ðšðððððð
ð¶ = ð% 141 79 95 141 105
ð¶ = ð% 56 31 38 56 41
With a sub exposure
value of 141sec it
possible to satisfy the
dominance criterion
for thresholds â¥2%
20. Dynamics criterion (1/2)
⪠The sky background signal induced by light pollution (which
represents the area of lowest intensity on each image) increases
with exposure time and therefore reduces the range of signal
collected (loss of dynamics range). We propose here a criterion for
determining a ð¡ððð¥ allowing to control the loss in dynamics.
⪠The dynamics range of an image in bits is calculated as follows:
ð· ð¡ = ððð2
ðððð¥
ðððð
20
Where ðððð¥ corresponds to the saturation (difference between the maximum value in ADU and the smallest recorded
signal value) and ðððð a minimum signal reference (most often we choose ðððð = ðð ð).
⪠In our case study we calculate the dynamics by the relation:
ð· ð¡ = ððð2
2ð â ð¡ à ððâ/ðŽð·ð à ðð¿ð
ðŽð·ð
ð¡ððð
2
/ð¡ððð
ðð ð
⪠Consequence : given a dynamics loss tolerance hypothesis, denoted by âð« (in bits), we can then determine
the maximum sub exposure time, denoted by ð¡ððð¥, through the image dynamics criterion.
The dynamics decreases with ð¡
21. Dynamics criterion (2/2)
⪠Thus, we define ð¡ððð¥ as follows:
ð· ð¡ððð â ð· ð¡ððð¥ = âð·
⪠To calculate ð¡ððð¥, we use the equation below:
ð¡ððð¥ =
2ð
â 2ð· ð¡ððð ââð·
à ðð ð
ððâ/ðŽð·ð à ðð¿ð
ðŽð·ð
ð¡ððð
2
/ð¡ððð
⪠Most often, experts recommend a value of âð« close
to 0.8 bits. Below the detail of the ð¡ððð¥ estimation:
21
Estimation of ðððð
ð¡ððð ð (*) 30
âð· (ððð¡ð ) 0.8
ðððð (ð) 225
(*) The value ð¡ððð considered is the
smallest admissible element (31 sec
exactly) among the values ââobtained
previously
23. Tracking error criterion on sub frames (2/2)
⪠The estimator á
ð of ð is defined as the empirical mean of the
estimators per session:
á
ð = â
1
ð
à·
ð=1
ð
1
ð¡ð
ðð ððð
⪠Below the results of estimation:
⪠By denoting ðŒ the maximum tolerated defection rate, we deduce the
maximum sub exposure time, denoted ð¡ððð, satisfying the criterion
ððâðð¡ððð ⥠ð 1 â ðŒ . This time is estimated by using the formula:
ð¡ððð = â
1
ð
ðð 1 â ðŒ
⪠For ðŒ = 10%, we have ð¡ððð = â
1
0,056%
ðð 1 â 10% , and
ðð ðð = ððð ððð
23
Lambda estimations
Average 0.056%
Median 0.055%
For greater caution,
we retain the value
à·
ð = 0.056%
26. Spectral band, energy et photons flux (2/2)
⪠In the following, for all 3 spectral domains, we will only refer to the flux density ððœ associated with the
visible domain, and we will consider a single bandwidth.
⪠The retained bandwidth (denoted by â) corresponds to the widths of the unitary spectral bands
(denoted by ð¿ðµððµ, ð¿ðµðð ððð ð¿ðµðð ) weighted by flux densities (denoted by ð¹ðµ, ð¹ð ððð ð¹ð ). We get:
â= ð¿ðµððµ Ã
ð¹ðµ
ð¹ð
+ ð¿ðµðð + ð¿ðµðð Ã
ð¹ð
ð¹ð
⪠From the assumptions considered, we have:
ââ 66 Ã
6.60085. 10â11
3.60994. 10â11
+ 102 + 141 Ã
2.28665. 10â11
3.60994. 10â11
â 312 ðð.
⪠We will retain in the following the parameterization: ââ ððð ðð.
26
27. Link between flux and magnitude
⪠We recall below the formula to evaluate the flux ðð of a sky object according to its magnitude ð
expressed in the Vega system (zero magnitude for Vega):
ðð = ððœ à ððâð/ð.ð
With ð¹ð = 3.60994. 10â11 ð/ð2/ðð.
⪠The elements ð¹ð and ð¹ð above correspond to point source flux which can be adjusted in surface
flux expressed for example in ð/ð2
/ðð /ððð ð ðð² without modifying the calculation formula.
27
29. Photon counting (2/3)
⪠We apply the previous formula to the Sky Background signal and to the object signal.
⪠Sky Background Brightness
⪠Suppose that the observation location where the image is taken has an SQM (Sky Quality Meter)
equal to 18.16 ððð / arcsec². The flux of photons per second associated with the Background
Sky (BS) signal ðð¿ð
ðâ
is as follows:
ðð¿ð
ðâ
= 9.98. 108 Ã 10â18.16/2.5 Ã
4.68
5
2
= ðð. ðð ððððððð / ð
⪠Object brightness
⪠The surface brightness of the target object (galaxy M33) is 23.25 ððð / arcsec². We
nevertheless prefer the peak luminosity(*) equal to 20.1 ððð / ððð sec² for the calculation of
the signal of the object. The flux of photons per second associated with the signal from the
object ðððð
ðâ
is as follows:
ðððð
ðâ
= 9.98. 108 Ã 10â20.1/2.5 Ã
4.68
5
2
= ð. ðð ððððððð / ð
29
(*) See the Messier catalog proposed by Tony Flanders: https://tonyflanders.wordpress.com/messier-guide-index-by-number/
30. Photon counting (3/3)
⪠The table below shows the empirical estimates of the number of photons per channel as well as the
theoretical estimates:
⪠The empirical and theoretical photon numbers associated with the signal from the object are very
close.
⪠Nevertheless, there is a discrepancy in the numbers of photons of the Sky Background signal which
is largely due to the SQM hypothesis. This assumption was not evaluated for the observation spot
but was extracted from the Light Pollution Map(*) site.
⪠As an example, an SQM of 18.5 ððð/arcsec² leads to an average number of 34.81 photons/s much
closer to the calculated empirical values. The sensitivity to this assumption is therefore particularly
substantial.
30
(*) https://www.lightpollutionmap.info/
Number of
photons
Channel R Channel G Channel B
Theoretical
calculation
ðð¿ð
ðâ
34.0 34.3 35.4 47.61
ðððð
ðâ
6.9 7.1 7.6 7.97
31. SNR value and profile (1/2)
⪠It is possible to exploit the previous photometry results to evaluate the signal-to-noise ratio of the
observation site ð» (for Home) where the image was made.
⪠By denoting ð¡ð» the total exposure time, ð¡ð¢ the unit exposure time, the signal-to-noise ratio is equal to:
ððð ðð»
(ð¡ð») =
ðððð
ðâ
ððâ/ðŽð·ð
ðððð
ðâ
ððâ/ðŽð·ð
2 +
ðð¿ð
ðŽð·ð
ð¡ððð
2
ð¡ððð
+
ðð·ð¶
ðŽð·ð
ð¡ððð
2
ð¡ððð
+
ðð ð
ðŽð·ð
²
ð¡ð¢
à ð¡ð»
With: ðððð
ðâ
= ððž à ð à ð ðµððŠðð à ð¶ à 10âðððð/2.5
Ã
ð(ðð)
ð
2
31
For ð¡ð¢ = 240ð , we have the following
signal-to-noise ratio : ðºðµð¹=ðð.ð
Remember that the total exposure time of
the case study is 5h20â
32. SNR value and profile (2/2)
⪠By setting a target SNR denoted ððð ð¡ðð¡
, the total exposure time ð¡ð¡ðð¡
required is calculated as follows:
ð¡ð¡ðð¡ = ð¶ð»
ððð ð¡ðð¡. ððâ/ðŽð·ð
ðððð
ðâ
2
Where,
ð¶ð» =
ðððð
ðâ
ððâ/ðŽð·ð
2
+
ðð¿ð
ðŽð·ð
ð¡ððð
2
ð¡ððð
+
ðð·ð¶
ðŽð·ð
ð¡ððð
2
ð¡ððð
+
ðð ð
ðŽð·ð
²
ð¡ð¢
32
Astrophotography experts often consider a target of ðºðµð¹ = ðð.
To achieve such an SNR, the exposure time must be 6h07' (instead of 5h20' in the case study).
33. Alternative approach of Read Noise dominance
⪠This alternative technique aims to improve the
previously detailed Read Noise dominance method,
by considering the entire signal-to-noise ratio as the
basis for calculation.
⪠We introduce ð =
ðððð
ðâ
ððâ/ðŽð·ð
2 + ðð¿ð
ðŽð·ð
ð¡ððð
2
Ã
1
ð¡ððð
+
ðð·ð¶
ðŽð·ð
ð¡ððð
2
Ã
1
ð¡ððð
⪠We solve the following inequality:
ðððð
ðâ
. ð¡/ððâ/ðŽð·ð
ð¡ð + ðð ð
ðŽð·ð
²
⥠1 + ðŒ
ðððð
ðâ
. ð¡/ððâ/ðŽð·ð
ð¡ð
⪠Solving the inequality leads to the following
solution ðððð
â
:
ð¡ððð
â
=
ðð ð
ðŽð·ð2
ð
Ã
1
1 + ðŒ 2 â 1
33
Threshold Alternative
method
Initial method
ð¶ = ð% 116 141
ð¶ = ð% 46 56
Values of ð¡ððð in sec (channel R)
SNR excluding
RN
34. Summary of sub exposure time constraints (1/2)
34
Maximum sub
exposure time
Minimum sub
exposure time
Minimum exposure time
associated with the Read Noise
dominance (alternative method
ðŒ = 5%)
ðððð = ðð ð
Minimum exposure time associated
with the Read Noise dominance
(alternative method ðŒ = 2%)
ðððð = ððð ð
Minimum exposure time
associated with the Read Noise
dominance (standard method
ðŒ = 2%)
ðððð = ððð ð
Maximum exposure
time associated with the
defection rate criterion
ðð ðð = ððð ð
Maximum exposure time
associated with the loss in
dynamics criterion
ðð ðð = ððð ð
Recommended exposure
time satisfying all the
criteria
ðð = ððð ð
Sub exposure time of
the case study
ððððð ðððð ð = ððð ð
35. Summary of sub exposure time constraints (2/2)
35
⪠The graph below displays the constraints induced by the dominance method, the management of the
defection rate and of the dynamics:
Range of
admissible sub
exposure times
36. Comparison of 2 observation sites SNRs (1/3)
⪠The practitioner may need to compare the total exposure time between an area where light pollution
is intense and a spot with a relatively pure sky background.
⪠We can for example ask the following question => What is the required exposure time (denoted ð¡ð»)
on an area of strong light pollution (called ð» for Home) to obtain the same SNR as that resulting from
a fixed exposure duration (denoted ð¡ðð) on a low polluted site (called ðð for Other Site)?
⪠Letâs introduce the SNRs of each of the sites:
ððð ðð»
ð¡ð» =
ðððð
ðâ
ððâ/ðŽð·ð ð¶ð»
à ð¡ð» ð€ðð¡â ð¶ð» =
ðððð
ðâ
ððâ/ðŽð·ð
2 +
ðð¿ð
ðŽð·ð
ð¡ððð
2
ð¡ððð
+
ðð·ð¶
ðŽð·ð
ð¡ððð
2
ð¡ððð
+
ðð ð
ðŽð·ð
²
ð¡ð¢
And
ððð ððð
ð¡ðð =
ðððð
ðâ
ððâ/ðŽð·ð ð¶ðð
ð¡ðð ð€ðð¡â ð¶ðð =
ðððð
ðâ
ððâ/ðŽð·ð
2 +
ðð
ðð¿ð
ðâ
ððâ/ðŽð·ð
2 +
ðð·ð¶
ðŽð·ð
ð¡ððð
2
ð¡ððð
+
ðð ð
ðŽð·ð
²
ð¡ð¢
With ðð» =
ð¡ð»
ð¡ð¢
and ððð =
ð¡ðð
ð¡ð¢
36
37. Comparison of 2 observation sites SNRs (2/3)
⪠In the previous expressions the signal of the object is
estimated by means of the relation:
ðððð
ðâ
= ððž à ð à ð ðµððŠðð à ð¶ à 10âðððð/2,5
Ã
ð(ðð)
ð
2
⪠Furthermore, it is assumed that the noise of sky
background (SB) is empirically calculated for ð» (on a
light frame) and is theoretically estimated for ðð from an
index ðºðžðŽð¶ðº = ðð. ðð mag / arc sec² with the formula:
ðððð¿ð
ðâ
= ððž à ð à ð ðµððŠðð à ð¶ à 10âððððð/2.5 Ã
ð(ðð)
ð
2
⪠We get the following equation of time:
ð¡ð»
â
=
ð¶ð»
ð¶ðð
à ð¡ðð
And the final result:
37
Intermediate elements Value
Signal (ADU / s) : ðððð
ðâ
/ððâ/ðŽð·ð 1.7
Variance Signal (ADU / s) : ðððð
ðâ
/
ððâ/ðŽð·ð
2 3.7
Variance SB (ADU / s) for ð» :
ðð¿ð
ðŽð·ð
ð¡ððð
2
/ð¡ððð
19.2
Variance DC (ADU / s) : ðð·ð¶
ðŽð·ð
ð¡ððð
2
/
ð¡ððð
0.48
Variance RN (ADU / frame) : ðð ð
ðŽð·ð
² 110
Variance SB (ADU / s) for ðð : ðð
ðð¿ð
ðâ
. ð¡ðð/
ððâ/ðŽð·ð
2 1.3
Sub exposure time (s) : ð¡ð¢ 150
Time ratio Value
Total variance (ADU / s) for ð» : ð¶ð» 24
Total variance (ADU / s) for ðð : ð¶ðð 6
Total exposure time for ðð (s) : ð¡ðð 1
Total exposure time for ð» (s) : ð¡ð»
â
3.86
ðð¯
â
â ð. ðð à ðð¶ðº
38. Comparison of 2 observation sites SNRs (3/3)
Comparison of times to reach a
target SNR:
38
Time required to achieve an SNR = 50
Site ðð: ð¡ðð 1h35â
Site ð»: ð¡ð» 6h07â
From an operational point of view, these orders of magnitude induce very different constraints. For
example, in the case of an astronomical night whose duration is less than 5h, it is impossible to
obtain an SNR ratio of 50 for the site ð¯ even though the latter is reached in 1h35 for ð¶ðº.
39. Sensitivity of the SNR to the signal of object value
39
⪠The calculations presented above derive from a theoretical
assumption of so-called "peak" luminosity equal to 20.1
mag/arcsec². This assumption leads to a flux of 7.97
photons/s, emanating from the brightest part of the target
object.
⪠In addition, we have also seen that the empirical
measurement of the signal in the core of M33 directly
carried out on a sub frame, enabled us to estimate a value
of 7.21 photons/s (average of the R, G and B channels).
⪠We present here a sensitivity analysis of the SNR to this
estimate.
Intermediate elements
Signal (ADU / s) - initial value:
ðððð
ðâ
/ððâ/ðŽð·ð
1.7
Variance Signal (ADU / s) - initial
value: ðððð
ðâ
/ððâ/ðŽð·ð
2 3.7
Signal (ADU / s) - empirical value:
ððð¡
ðððð
ðâ
/ððâ/ðŽð·ð
1.5
Variance Signal (ADU / s) - empirical
value: ððð¡
ðððð
ðâ
/ððâ/ðŽð·ð
2 3.4
Time required to achieve an SNR = 50
ð¡ð» initial value 6h07â
ð¡ð» empirical value 7h20â
41. Photons number law of probability: introduction
⪠It is most often assumed that the number of photons emitted by astronomical objects or the signal from
the Sky Background follows a Poisson process. In general, two arguments are developed to justify this
phenomenon.
⪠The first consists in considering a set of hypotheses similar to the mathematical definition of the Poisson
process. Below, these assumptions:
⢠The probability of detecting a photon over an infinitesimal time interval is proportional to an intensity parameter:
ð ðð¡+âð¡ â ðð¡ = 1 = ðâð¡ + ð âð¡ ;
⢠The probability of detecting a quantity greater than one photon is negligible: ð ðð¡+âð¡ â ðð¡ > 1 = ð âð¡ ;
⢠Photon detections associated with disjoint time intervals are statistically independent.
⪠Under these assumptions, the number of photons emitted by the target object follows a Poisson process.
In mathematics, a Poisson process is exactly defined by the 3 hypotheses formulated above.
⪠An alternative strategy consists in showing that the poissonian distribution of light comes from intrinsic
properties of electromagnetic radiation. Thus it is possible (under good assumptions) to demonstrate using
quantum mechanics results associated with the quantum harmonic oscillator that the number of photons
is indeed Poisson distributed.
41
42. Photons number law of probability: quantum mechanics (1/3)
⪠In quantum optics, a constant intensity light source can be
modeled as a coherent state of a quantum harmonic
oscillator.
⪠We recall the expression of the quantum Hamiltonian:
ð» =
1
2
âð ð2 + ð2
With
ð =
ðð
â
à·
ð ðð¡ ð =
1
ððâ
Æž
ð
⪠We define the annihilation and creation operators denoted
ð and ðâ respectively as follows:
ð =
1
2
ð + ðð ðð¡ ðâ
=
1
2
ð â ðð
⪠The number operator, denoted ð, verifies: ð = ðâð
⪠It is easily shown that: ð» = âð ð +
1
2
42
Theorem:
(i) The spectrum of ð is â. It follows that the
spectrum of ð» is ð +
1
2
âðÈ ð â â
(ii) If Û§
Èðð is an eigenvector of ð associated
with the eigenvalue ð, then ðâ Û§
Èðð is an
eigenvector of ð associated with the
eigenvalue ð + ð
(iii) If Û§
Èðð is an eigenvector of ð associated
with the eigenvalue ð, then if ð â 0, ð Û§
Èðð
is an eigenvector of ð associated with the
eigenvalue ð â ð and if ð = 0, ð Û§
Èð0 = 0
43. Photons number law of probability: quantum mechanics (2/3)
⪠Note that the eigenstates of ðµ and ð¯ are identical. These states are also called Fock states.
⪠We show that if ðð is an eigenstate of ðµ - or equivalently of ð» - associated with the eigenvalue ð or
ðžð = ð +
1
2
âð then:
ðð ð¢ =
1
ð
1
4 1
2ðð!
ðâ
ð¢2
2 ð»ð ð¢
With ð»ð ð¢ the Hermite polynomial of order ð.
ð»0 ð¢ = 1, ð»1 ð¢ = 2ð¢, ð»ð+1 ð¢ = 2ð¢ð»ð ð¢ â 2ðð»ðâ1 ð¢ ð ⥠2.
⪠Most often we denote ۧ
Èð â¡ ðð, the Fock state associated with the eigenvalue ð of the operator ð
and the eigenvalue ðžð = ð +
1
2
âð of the operator ð».
43
It is assumed that the system is in a coherent state. These states, also called quasi-classical states,
correspond to the purest states that can be defined. In particular, they minimize the uncertainty
principle and come close to the solutions of the classical harmonic oscillator.
44. Photons number law of probability: quantum mechanics (3/3)
⪠Coherent states are defined as the eigenstates of the ð operator. Let ð§ â â, so we have: ð Û§
Èð§ = Û§
ð§Èð§
⪠And we can prove the following formula: ۧ
Èð§ = Ïð
ð§ð
ð!
ðâ ð§ 2/2 Û§
Èð
⪠If the system is in the coherent state ۧ
Èð , then we can determine the probability associated with the energy
value ð¬ð = ð +
ð
ð
âð. This probability, denoted ð·ð ð , corresponds to the squared modulus of the
projection on the Fock state associated with ðžð. So we have :
ðð ð§ =
ð§ð
ð!
ðâ ð§ 2/2
2
=
ð§ 2 ð
ð!
ðâ ð§ 2
⪠We get the elementary probability of a Poisson distribution (with parameter ð§ 2) associated with the count
of ð occurrences.
⪠Under the usual notations of quantum mechanics, we can therefore write that:
ð Û§
Èð§ = ð§ 2
ðð¡ âð Û§
Èð§ = ð§ 2 = ð§
44
We conclude that the number of photons measured from a coherent state follows a Poisson
distribution. An immediate consequence is that, over a fixed time interval, the expectation and the
variance of the number of observed photons are identical.
45. Equality expectation / variance of the photoelectrons number (1/2)
⪠The number of photons reaching the sensor is therefore assumed to follow a Poisson distribution. As an
example, consider ðððð
ðâ
ð¡ the process associated with the number of photons emitted by the target object.
So we have :
ðž ðððð
ðâ
ð¡ = ð ðððð
ðâ
ð¡ = ðððð
ðâ
. ð¡
⪠Note: expectation and variance are identical and they are linear in ð.
⪠The number of electrons generated by the sensor is not exactly equal to the number of photons reaching it.
The average numbers of electrons and photons differ by a factor called Quantum Efficiency, noted ððž.
More specifically we have:
ððž =
ðž ðððð
ðâ
ð¡
ðž ðððð
ðâ
ð¡
⪠Another way to formalize it is to assume that when a photon hits the photosite it generates an electron
with probability ðžð¬ and produces nothing with probability ðâðžð¬. This can be modeled as follows:
ðððð
ðâ
ð¡ = à·
ð=1
ðððð
ðâ
ð¡
ðð
45
Where ðððð
ðâ
ð¡ follows a Poisson
process and ðð a Bernoulli distribution
with parameter ððž.
46. Equality expectation / variance of the photoelectrons number (2/2)
⪠Under such a model (called compound Poisson), we can therefore calculate the expectation and variance
of the random variable ðððð
ðâ
ð¡ :
ðž ðððð
ðâ
ð¡ = ðž ðž ðððð
ðâ
ð¡ ðððð
ðâ
ð¡ = ðž ððž Ã ðððð
ðâ
ð¡ = ðžð¬ à ðððð
ðð
. ð
And,
ð ðððð
ðâ
ð¡ = ðž ð ðððð
ðâ
ð¡ ðððð
ðâ
ð¡ + ð ðž ðððð
ðâ
ð¡ ðððð
ðâ
ð¡
= ðž ðððð
ðâ
ð¡ Ã ððž Ã (1 â ððž) + ð ððž Ã ðððð
ðâ
ð¡
= ððž. 1 â ððž . ðððð
ðâ
. ð¡ + ððž2
. ðððð
ðâ
. ð¡
= ððž. ðððð
ðâ
. ð¡ Ã 1 â ððž + ððž
= ðžð¬. ðððð
ðð
. ð
= ðž ðððð
ðâ
ð¡
46
In conclusion, even though the
electron counting process is not
Poisson, it can be shown that its
expectation is equal to its
variance and that they are linear
in ð.
48. Bibliographie (1/2)
Conference by Robin Glover (SharpCap software developer) - Practical Astronomy Show 2019
Deep Sky Astrophotography With CMOS Cameras
https://www.youtube.com/watch?v=3RH93UvP358&list=WL&index=46
Didier Walliang (2018) - Le bruit en astrophotographie
RCE presentation November 2018
https://media.afastronomie.fr/RCE/PresentationsRCE2018/Walliang-RCE2018.pdf
Thierry Legault - Astrophotographie
Editions Eyrolles
Lecture material on Computational Imaging and Sensor Theory (Stanford University)
https://isl.stanford.edu/~abbas/ee392b/lect08.pdf
Henry Joy McCracken (2017) - Institut dâAstrophysique de Paris
An introduction to photometry and photometric measurements
http://www2.iap.fr/users/hjmcc/hjmcc-photom-ohp-2017.key.pdf
48