Quantum efficiency graph of MN34230 (ASI1600 / QHY163 / Atik Horizon)

[vlaiv]

I'm doing some research into colorimetric calibration of images and need precise QE graph of above sensor.

Internet is flooded with same graph that goes from 400 to 800nm and I suspect it's general copy of the same thing repeating all over the place, however Christian Buil did comparison of three camera models using this sensor (Atik Horizon and two ASI models MM and PRO), and his page provided measured QE graph for ASI1600MM (above Panasonic sensor). Here is mentioned report:

http://www.astrosurf.com/buil/atik_vs_zwo/

I've transferred both graphs into csv file using this handy tool for that:

https://apps.automeris.io/wpd/

(WebPlotDigitizer)

and imported data into spreadsheet. However, once I normalized both graphs (divided with max value for each) - I got two completely different things - different curves. Here is result:

Blue (column F) - is "standard" internet based QE graph, while column G / orange one is Buil's curve. There is obviously large discrepancy between the two. It's not due to scale factor (I don't think I messed up anything with scaling) - curves look distinctly different - neither horizontal nor vertical scaling (frequency problem, or magnitude problem) can make them overlap.

Anyone has idea what might be going on here, or even better, does anyone have "reliable" QE graph for this sensor (or value table / data), preferably in 380-780nm range (internet published one starts at 400nm)?

 

[Thalestris24]

Can't help other than it's been discussed before here. Seems the peak qe is only about 60% which is maybe why the manufacturers don't shout about it!

Louise

[vlaiv]

14 minutes ago, Thalestris24 said:

Can't help other than it's been discussed before here. Seems the peak qe is only about 60% which is maybe why the manufacturers don't shout about it!

Louise

Hi Louise,

Thanks for that, unfortunately, peak QE is not something that will help me - I'm after exact shape of the curve, even in relative form. For color calculations, it is relative response per frequency that is important.

[Thalestris24]

9 minutes ago, vlaiv said:

Hi Louise,

Thanks for that, unfortunately, peak QE is not something that will help me - I'm after exact shape of the curve, even in relative form. For color calculations, it is relative response per frequency that is important.

Sorry. I'd have thought that if the info can't be found via Google, then it's maybe not in the public domain, other than what you already have. If it comes to it, I guess you could try contacting Panasonic directly.

ps Am I experiencing deja-vu re this question?

Louise

[vlaiv]

1 hour ago, Thalestris24 said:

Sorry. I'd have thought that if the info can't be found via Google, then it's maybe not in the public domain, other than what you already have. If it comes to it, I guess you could try contacting Panasonic directly.

ps Am I experiencing deja-vu re this question?

Louise

Not sure, I haven't ask this one before, I just took graph provided by ZWO and other as accurate one prior to this. Problem emerged because I needed 380-780nm range, and that graph provides 400-800nm range, and it's obvious from the graph that QE does not fall sharply to 0 below 400 - so I set out to find another graph that might provide this info when I stumbled upon page by Buil and found significant difference between published graph and his measurement.

I'm just fiddling now with the data and it seems that one of graphs is not corrected for photon flux / energy flux relationship - meaning someone possibly used reference curve in energy flux and calibrated photon flux sensor response with it.

[vlaiv]

Hm, yes, this is much better match, but I'm still not certain what I would accept this much error in measurement.

Above graph is: Blue line / column P is original ZWO published QE curve.

Orange line / column Q was obtained by following process: I first corrected Buil's curve by multiplying values by wavelength (energy of a photon is hc/lambda so we multiply by lambda to get rid of wavelength dependence and get pure photon count) and then did least squares fitting (curves don't have common peak to do simple peak scaling).

This, unfortunately, does not give me a clue which one of the graphs is accurate (or less wrong ).

[Thalestris24]

Oh, I just had a feeling you'd asked before and that I'd replied as I did. Just me being weird - else a glitch in 'the matrix' ha ha.

There seem to be many variables with these cmos cameras and probably variation between different models using the same sensor. Maybe someone has done some possibly useful measurements in the spectroscopy section - just a wild thought!

Louise

[vlaiv]

7 minutes ago, Thalestris24 said:

Oh, I just had a feeling you'd asked before and that I'd replied as I did. Just me being weird - else a glitch in 'the matrix' ha ha.

There seem to be many variables with these cmos cameras and probably variation between different models using the same sensor. Maybe someone has done some possibly useful measurements in the spectroscopy section - just a wild thought!

Louise

Yes, that was my next line of reasoning - how to perform measurement myself.

It sounds quite straight forward - take Star Analyzer, take spectrum of known star, divide measured one with reference - and "voila" we have sensor response. Unfortunately it's not that easy. One needs to account both for instrument response (mirrors don't have equal reflectivity with respect to wavelength, I wonder if refractive optics fares better in this regard - transmission might be equal, or at least with much less variation), and also atmosphere changes spectrum based on how much transparency there is. I'm not sure if frequency response of atmosphere varies with conditions or if it's always linear with respect to transparency (for example different gasses attenuating wavelengths differently, or transparency affects all wavelengths equally).

Best approach that I can think of would be "laboratory setup" - which requires broadband source of known spectral distribution / pinhole (maybe optic cable - these are quite narrow, about 10um or so) and SA mounted on refractor (to avoid mirrors).

[Thalestris24]

5 minutes ago, vlaiv said:

Yes, that was my next line of reasoning - how to perform measurement myself.

It sounds quite straight forward - take Star Analyzer, take spectrum of known star, divide measured one with reference - and "voila" we have sensor response. Unfortunately it's not that easy. One needs to account both for instrument response (mirrors don't have equal reflectivity with respect to wavelength, I wonder if refractive optics fares better in this regard - transmission might be equal, or at least with much less variation), and also atmosphere changes spectrum based on how much transparency there is. I'm not sure if frequency response of atmosphere varies with conditions or if it's always linear with respect to transparency (for example different gasses attenuating wavelengths differently, or transparency affects all wavelengths equally).

Best approach that I can think of would be "laboratory setup" - which requires broadband source of known spectral distribution / pinhole (maybe optic cable - these are quite narrow, about 10um or so) and SA mounted on refractor (to avoid mirrors).

Oh well, good luck! I guess manufacturers have quite sophisticated setups for such measurements - or maybe not!

[Melitastro]

Something that comes to mind when mentioning errors is that these graphs don't show data points, and along with data points you normally expect to see error bars on them. I have no idea how many variables there are when performing these measurements but as an example how accurate is the temperature? If my camera states that it's cooled to -20°C I have no way of verifying it.  It may help to find how much uncertainty there are on these measurements and then see how the graphs compare?

[drjolo]

I have only compared sensitivity of my QHY163M camera to Atik383 sensor using HaLRGB filters on the same starfield. Here is the result table:

It is little bit less sensitive in R (than KAF8300, but then for G and B it becomes more sensitive. No more data  

[vlaiv]

3 hours ago, Melitastro said:

Something that comes to mind when mentioning errors is that these graphs don't show data points, and along with data points you normally expect to see error bars on them. I have no idea how many variables there are when performing these measurements but as an example how accurate is the temperature? If my camera states that it's cooled to -20°C I have no way of verifying it.  It may help to find how much uncertainty there are on these measurements and then see how the graphs compare?

QE shouldn't depend on temperature and random error in measurement is just related to signal strength or time, so it can be arbitrarily small. Both graphs are quite smooth so very good SNR is achieved in both measurements.

There is obvious systematic error in one of the measurements (photon flux which sensor measures vs energy flux which is often used for standard sources) - which is one of the basis to question one of above results.

Published ASI1600 QE graph could be theoretical one, and that would somewhat explain discrepancy as actual QE depends on manufacturing process, but that would imply that Buil's measurement is flawed with systematic error - and I don't think he would make such an error. This page has been brought to my attention (also his measurements of QE graphs for other cameras):

http://www.astrosurf.com/buil/isis/noise/result.htm

and published and measured results are in very good agreement (there is some residual error - but at level that I would expect / be ok with).

[Thalestris24]

Hi Vlaiv,

Have you done any more on this? I see Christian Buil has had a look at a ASI183mm . Which is useful. He really doesn't like the long exposure glo! But I guess most people would do lots of much shorter exposures and calibrate the glo out.
Having just finished my uni classes, I'm having another look at better/sensitive cameras for (mostly) eaa/live stacking again. Was still thinking about the 183mm... However, I get seduced by the quoted high qe, but... the peak is shifted towards the blue-green part of the spectrum and is down to around only ~50% at the red end. Why can't someone make an ideal eaa camera that I can afford?? Ha ha. I'm still looking...

Louise

[andrew s]

Using a star give a combined instrument response and that of the atmosphere who's response is not insignificant.  Even  reflective optics will have an effect as will the grating. It is not easy.

Buil is a very respected and careful worker I  this area,  I would trust his results.

Regards Andrew 

[vlaiv]

4 hours ago, Thalestris24 said:

Hi Vlaiv,

Have you done any more on this? I see Christian Buil has had a look at a ASI183mm . Which is useful. He really doesn't like the long exposure glo! But I guess most people would do lots of much shorter exposures and calibrate the glo out.
Having just finished my uni classes, I'm having another look at better/sensitive cameras for (mostly) eaa/live stacking again. Was still thinking about the 183mm... However, I get seduced by the quoted high qe, but... the peak is shifted towards the blue-green part of the spectrum and is down to around only ~50% at the red end. Why can't someone make an ideal eaa camera that I can afford?? Ha ha. I'm still looking...

Louise

Hi Louise,

I did not pursue this further. At the time, I was interested in doing color gamut calculations for different sensors, but since then I figured out that it would take too much both my time to write such simulation and simulation to run with significant precision (at least brute force approach - dividing 400nm to 700nm into small chunks with let's say 10nm spacing and 0-100% light intensity range by 5% would need 20 to the power of 30 calculation - and that is ~ 1 followed by 39 zeros - clearly not best way to deal with the problem).

I also figured out that I'm going to do relative colorimetry - measuring filter values (basic photometry) and doing linear transform to XYZ space based on plate solve / catalog star color indices or stellar class data.

[Thalestris24]

3 minutes ago, vlaiv said:

Hi Louise,

I did not pursue this further. At the time, I was interested in doing color gamut calculations for different sensors, but since then I figured out that it would take too much both my time to write such simulation and simulation to run with significant precision (at least brute force approach - dividing 400nm to 700nm into small chunks with let's say 10nm spacing and 0-100% light intensity range by 5% would need 20 to the power of 30 calculation - and that is ~ 1 followed by 39 zeros - clearly not best way to deal with the problem).

I also figured out that I'm going to do relative colorimetry - measuring filter values (basic photometry) and doing linear transform to XYZ space based on plate solve / catalog star color indices or stellar class data.

Ok, no worries, I just wondered.