Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

ASI294MC Calibration – Testing, Notes, Thoughts, and Opinions

  • Please log in to reply
35 replies to this topic

#26 jdupton

jdupton

    Surveyor 1

  • *****
  • topic starter
  • Posts: 1780
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 14 October 2018 - 08:15 PM

Jon,

 

I wonder what experimentation would show here. With the drifting mean ADU values, I wonder if somehow camera self-calibration might be feeding off itself, and adjusting to a zero-photo signal situation. I wonder if, with a photographic signal present, if any self-calibration might be more consistent...

 

   I had already done a few informal experiments on this concept when I did my Flats. When I first noticed that mean ADU values differed between sequences with both Bias and Dark Frames (or Flat Frames) when compared to sequences which only captured Dark or Flat Frames, I started adding Bias Frames between all frame captures in my sequences. (Of course the Bias-like frames between the Flats were not true to form since the flat illumination source was still on. These were 0 second duration Flats rather than a real Bias but they acted the same in the sequence results.)

 

   This all came about originally as I was running sequences of only Dark frames to measure Dark Current for the camera. I ran sequences of Dark Frames between 15 seconds and 960 seconds in length at different temperatures. I found that if I plotted Dark Current (calculated from the slope between two points), I would get seemingly random variations of Dark Current at the shorter exposure times. When I ran sequences of Dark Frames again but with Bias Frames between each Dark Frame, The random variation seemed to go away.

 

   (Unfortunately, the plots then showed a very repeatable systematic trend at the same lower Dark Frame Exposures. This is still on my list for further investigation along with more study of session to session differences when Bias Frames are not used.)

 

 

John


Edited by jdupton, 14 October 2018 - 08:17 PM.


#27 ac4lt

ac4lt

    Ranger 4

  • -----
  • Posts: 384
  • Joined: 16 Aug 2017
  • Loc: Northern Virginia

Posted 18 October 2018 - 08:15 PM

I've been following this thread with interest and I decided to see how my ASI294 was behaving. I tried to replicate John's method as best I could. I set the ASI driver to unity gain and in SGP set the temperature to 0C. I let the temperature stabilize for at least 5 minutes before running the sequence. I used the same times and number of exposures. Once captured I ran integrations in PixInsight and then used the statistics process to get the mean and then scaled that by 65, 535 to convert back to ADU. Perhaps I should have scaled by 14 bits rather than 16, but for comparison purposes this should be ok. I got a graph that looked very much like John's though with different but similar values (probably because of how I scaled).

 

I decided to run another sequence with a 3 second delay between exposures. I though that if it was heat related as John expected then this might give some of that generated heat a chance to dissipate and ADU values would be lower. And this seemed to be correct, except for the bias frame with the 0 length exposure. It came out almost exactly the same as the sequence with no delays.

 

Here's the graph for the two sets:

Exposure (sec) vs. ADU.png

 

It makes sense to me that the values on the 3-second delay sequence are lower but I'm a bit puzzled by the unchanged bias value. The only thing I can think of is that if the circuitry that is causing the extra heat is only on for a part of the exposure time then it gets less effective time to dissipate on the bias frames. That might explain why the graph trends down so quickly where it trends up quickly on the no-delay sequence. In that case (no-delay) the heat builds up over the course of the sequence.

 

I suppose one could plot the ADU values for the individual exposures and see if they trend up in the no-delay sequence and stay closer to stable in the 3-second-delay sequence. That's a bit more manual data mining than I can do right now. Is there a way to feed a set of files to PixInsight and have it spit out the median for each one to a file or the screen as a list?

 

Versions:

SGP 3.0.2.94

ASI Native Driver: 3.0.0.3

ASI Ascom Driver 1.0.3.22

Ascom Platform: 6.4.2.2618

 

I'm not sure what value this extra info has but wanted to share it in case it helped.

 

After reading through the thread it sounds like there are these different ways to cope:

1. Don't take bias and just shoot darks and dark-flats

2. Rescale the master bias according to the process John described (finding the intercept of the 3s-6s line) and shoot darks normally and keep flats over 3 seconds

 

Have I got that right? If using darks for flats then there is no reason to need to shoot long flats since the dark flats would match the same, right? Though it may still be desirable to try to use longer flats.

 

Am I missing something or is that a fair summary?


  • h2ologg and jdupton like this

#28 jdupton

jdupton

    Surveyor 1

  • *****
  • topic starter
  • Posts: 1780
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 18 October 2018 - 09:36 PM

Linda,

 

   Here are some comments out of order from your questions and comments.

 

   The extra test data you generated has a lot of value. It is always good to have others duplicate an experiment to make sure the first person (me) wasn't just generating garbage. Thanks for the time you put into this.

 

   I think your conclusions after running the experiment match my own. Recall, that you don't have to shoot Bias Frames at all unless you may want to scale your Dark Frames for some other reason. So long as you shoot for Flats longer than about 3 seconds, your Flat-Darks should be all you need for calibration. Just staying out of the exposure zone of less than 3 seconds should suffice.

 

   Regarding getting the data out of PixInsight more easily, it is drop dead easy. Try this procedure:

  • Open the Blink Process in PixInsight
  • Load up all of your experiment frames into Blink. You can load Bias, Dark, Flat, and even Light frames all at once if you wish. It makes everything available for analysis.
  • Select the first Frame in the Blink Process file list.
  • Use Ctrl-A to select all the rest of the files.
  • Click the next to last icon at the bottom of the Blink window. (Series Analysis Report) This will output the basics but you should also output some additional data while you are at it. It can allow you to analyse many other aspects of the data.
  • In the "Statistics" window that opens from the Blink Process, select the following additional data (using a check mark in the box) to output in the report: 
    OBJECT         (This is the target name from SGP's sequence)
    DATE-OBS     (UTC time of frame capture)
    IMAGETYP     (Image Type -- Dark, Flat, etc)
    EXPOSURE    (Exposure time -- this will be repeated in the report. some PI processes set the other to 0)
    CCD-TEMP     (The actual temperature reported for the frame at time of capture)
    SET-TEMP      (The set-point temperature you requested. This can and often is different than above)
    GAIN               (The gain setting from the driver)
    EGAIN            (The electron gain assumed by the driver)
  • After selecting the above extra data, check the Write Text File Box 
  • Select the output folder for the report. It will always be called "Statistics.txt"
  • Press OK to generate the report.

   At that point, blink will grab all the data from every file and put in the Statistics.txt report file. Now you can load all that into a spreadsheet. I use LibreOffice Calc but you can use Excel or OpenOffice or most any other spreadsheet. To look at and plot the data, I use this method:

  • Open the spreadsheet.
  • Use File | Open to open the statistics.txt file.
  • You will get the import dialog. I just take the defaults and press OK.
  • You will then get a listing of all the statistics and extra information with each frame on a separate line and all the data lined up in proper columns.
     
  • I do the analysis using a pivot table. It is the easiest, quickest method to get a summary of all the data.
     
  • Select all the data on the sheet -- just click in the upper left corner for all lines and columns
  • Select Data | Pivot Table | Create from the menus
  • Use "Current Selection" (the default) in the Select Source dialog that pops up.
  • You now get the Pivot Table Layout dialog
     
  • Fill out the dialog as described below.
     
  • Click twice on the highlighted "Data" designation in the Column Fields area and press delete to remove it.
  • In the Available Fields area, drag the IMAGETYP data component from the list into the Page Fields area at the top and drop it there.
  • In the Available Fields area, drag the SET-TEMP data component from the list into the Columns Fields area and drop it there.
  • Drag the EXPOSURE data component from Available Fields to the Row Fields area and drop it there.
  • Drag the Mean data component from the Available Fields area to the Data Fields area and drop it there.
  • The Mean that you just dropped will now read "Sum - Mean". You will need to change that next.
  • Double click on the Sum - Mean data component and the Field Data dialog will pop up.
  • Select "Average" and press OK. The Data Fields area will now read "Average - Mean"
  • Expand the "Options" area of the Pivot Table Layout dialog.
  • Un-check the Total Rows and Total Columns check-boxes.
  • Press OK and the pivot table will be created in a new sheet. You can create it in a specific area but for this quick intro having it on a new sheet is just as good.

    The table will be created on a new sheet. It will contain a selector called IMAGETYP at the top where you can select just the types you want to see. For this exercise, use just the BIAS and DARK types. Next there will be a table which contains the table tile Average - Mean and a selector for SET-TEMP. The SET-TEMP will only contain a single entry if all the frames has the same Set-Point. Next down, there will be a selector for EXPOSURE. Make sure all are selected. The first data column has all the different  exposure times for all frames of the selected type. The second column has the Average of the Mean ADU for all frames of the selected types at the corresponding exposure time.

    The two data columns (Exposure and Mean ADU) can be plotted to give a result just like what you posted.

   I find it interesting that your second set of exposures with delay looks as it does. I am not totally surprised, though. If I might suggest an experiment for you try next time you have spare time, it would be to repeat that second part of the test but put dummy Bias Frames between the Dark Frame exposures and then have SGP rotate through the events. I think you may find that the lowered value of the mean ADU will rise back and begin to look like the first plot of frames without a delay in between. 

 

   I suggest this because that data is reminiscent of what I was seeing as inconsistent results in my own runs. Those results cleaned up for me when I placed Bias Frame captures between the Dark Frames.

 

   Another experiment to try is the run the the low exposure times in reverse order. In other words, instead of running the frames as 0.0, 0.2, 0.4, 0.6, 0.8, and 1.0, try them as 0.0, 1.0, 0.8, 0.6, 0.4, and 0.2. I expect you will get the same plot as your first one (in red). If so, it says that the steep slope is not due to cumulative self warming of the chip but is a result of the active timing circuitry on the chip.

 

[EDIT]   I say this because my original plot which looks like your No_Delay plot was actually run with a 10 second delay between all exposures. Nearly all of my data had the 10 second delay built into the SGP sequence.

 

   Anyway, thanks so much for going to the trouble of running this data through your own camera. It really does help to see what others are getting and bolsters my opinion that what I see from my camera is not unique.

 

 

John


Edited by jdupton, 19 October 2018 - 07:18 AM.


#29 jdupton

jdupton

    Surveyor 1

  • *****
  • topic starter
  • Posts: 1780
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 19 October 2018 - 01:33 PM

Linda and Others wishing to duplicate some of these experiments,

 

   Just for reference, here is a screen shot of my SGP sequence set up. I used similar sequences for all of my later testing after seeing the oddly inconsistent results of just shooting frames as I always had with my Mono CCD camera.

 

   You can note in this sequence, that in addition to inserting 0 second exposures between all Dark Frames, I also had the exposure values all mixed up just in case there might be some influence of shooting increasing very short exposures in order. The results for 0.2 to 0.8 seconds still had the very linear ramp demonstrating (in my mind) that there was not a cumulative warming effect on the sensor of repeated short duration increasing exposure times. Also note that I had already standardized on shooting all test frames with 10 second delays between frame exposures. The delay between exposures should have also ensured recovery time for any self-heating of the sensor.

 

   Rotating through all events rather than shooting the entire events in order ensured that every Dark Frame was preceded by a Bias Frame (of 0 exposure).

 

   Here is the sequence that generated the data plotted in Figure 1 of the first post in this thread.

 

Bias_Tests_+00_Capture-rsz.png

Fig 11 - Typical SGP Sequence for running experiments for jdupton's ASI294MC

 

 

John



#30 MARA25

MARA25

    Sputnik

  • *****
  • Posts: 28
  • Joined: 24 Sep 2017
  • Loc: Washington DC

Posted 21 April 2019 - 09:52 AM

Sorry for posting in an old thread. I have a question that may not deserve its own thread. I liked the discussion here and was wondering why the bias level is ~1900 adu?
From what I gathered the asi294mc is 14bit with a defaut offset of 50 and only 1 or 2 e- of read out noise. The 16bit adu value should then be on the order of 4x50=200 adu. What am I missing?

#31 jdupton

jdupton

    Surveyor 1

  • *****
  • topic starter
  • Posts: 1780
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 21 April 2019 - 10:10 AM

MARA25,

 

   That is a good question. I do not know the real answer. A couple of points come to mind.

 

   First, the actual A/D resolution of the camera is 14 bits. The camera's driver converts that to 16 bits by simply multiplying by four. Thus you might think the bias offset could be around ~475 ADU (~1900 / 4).

 

   For my camera and the ASCOM driver which I use, the default unity gain and offset are 120 gain and 30 offset. I have no direct knowledge of how the 30 offset number relates to the ~475 ADU output by the camera for a bias style frame. Somewhere in the camera, the gain value is translated into an offset voltage that is then measured by the A/D conversion. Having never seen a detailed specification for the sensor, I don't have any idea exactly how that relationship between offset setting value and voltage offset is determined. It is a black box for the most part and every camera manufacturer using this sensor may be doing it differently.

 

   Maybe someone who has knowledge of the internals of the sensor / camera(s) will have a better answer.

 

 

John


Edited by jdupton, 22 April 2019 - 08:12 AM.


#32 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23369
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 22 April 2019 - 03:44 PM

Sorry for posting in an old thread. I have a question that may not deserve its own thread. I liked the discussion here and was wondering why the bias level is ~1900 adu?
From what I gathered the asi294mc is 14bit with a defaut offset of 50 and only 1 or 2 e- of read out noise. The 16bit adu value should then be on the order of 4x50=200 adu. What am I missing?

I would be surprised if the offset was only 50 ADU. Usually 12-bit cameras have an offset of 50 ADU at higher gains, which at those gains is only a few electrons (barely enough).

 

The ASI294 IS a 14-bit camera, so ideally the offset would be at the very least 4x a large as 12-bit cameras. That would mean it should be at least 200 ADU (14-bit). That said, 50 ADU is a bit tight for higher gains on 12-bit cameras, and the 294 may have higher dark current and does have some brighter glows, in which case a larger offset could be necessary...a 400-500 ADU (14-bit) offset may be entirely appropriate. The scale factor between 14- and 16-bit is 4x, and a 400-500 ADU offset brings you to 1600-2000 DN (16-bit). 

 

Some Sony sensors also use an arbitrary numeric scale for their sensor offsets. The IMX183 for example uses a 5x scale. So an "offset" of 10 actually means 50 ADU. I don't know why they do this, it is kind of annoying that they do, but it is very likely the IMX294 also does the same thing. 


Edited by Jon Rista, 22 April 2019 - 03:46 PM.


#33 PeteM

PeteM

    Vanguard

  • -----
  • Posts: 2218
  • Joined: 12 Jul 2009
  • Loc: West Michigan

Posted 25 June 2019 - 09:54 AM

Sorry to dredge this up, but I do have a question in regards to the "spacer" bias/dark frames when doing flats. With the spacer frame being exposed to the flat panel and collection light, does that effect the purpose of the spacer? Or was it more about letting the sensor cool down a bit. If so, wouldn't the delay be enough? I do think in SGP if the flat panel is being controled via SGP, that it will turn it off for the dark or bias and then back on for the flat. I need to test that out tonight with my concern being the flat panel not being at the selected brightness level when the flat begins causing some issues.



#34 jdupton

jdupton

    Surveyor 1

  • *****
  • topic starter
  • Posts: 1780
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 25 June 2019 - 12:19 PM

Pete,

 

   No problem with using this thread. I think your question is a good follow-on to some of the recommendations I have made regarding use of the ASI294MC-Pro camera.

 

   Regarding light falling on the sensor during the spacer frames, yes, that does have an effect. In fact, recent testing I did for discussion in another thread here on CloudyNights, showed the reason for inserting dummy frames between exposures. See the following (rather long but useful) thread:

https://www.cloudynights.com/topic/662887-not-rbi-on-cmos-but-this-looks-similar/

 

   In that thread we pretty well proved that the sensor does not always completely reset the charge in pixel wells at the beginning of an exposure. There is an effect that another user referenced known as "CMOS Image Lag" which behaves similarly to RBI (residual bulk image) in Deeply cooled CCD sensors. The residual charge left in the pixel wells was able to "pollute" the next exposure. Doing a very short dark exposure effectively allows a second reset to happen before your next desired frame.

 

   For my camera, this effect was noted in about 10% of all exposures. My initial findings in this thread last year showed that the dummy frame captured between normal frames was effective at giving consistent results but it was not until the testing for that other thread that I understood why the dummy frames worked. My recent tests showed that either 0 second Bias Frames or 1 second Dark Frames between other exposures resulting in elimination of the residual charges. My testing indicated that a delay (in the dark) also seemed to help but I do not fully trust that conclusion.

 

   The OP, Andy, in the other thread discovered that just having light falling on the sensor between exposures could create the residual charge (image) some small percentage of the time. That means that when taking Flat Frames, you could have this show up unless the light source can be turned off and a Bias Frame taken between them. A simple delay with the light on between Flat Frames means that some frames will have extra residual light in them from the prior exposure to light (even during a delay).

 

   There are two pieces of good news, though. In Sequence Generator Pro, the Flat Frame light source (if controlled via SGP) is turned off during Dark and Bias Frames. That is how I tested my camera for this residual image effect. (See my test method described in the other thread referenced above.) So using SGP with a compatible Flat Frame Light Controller allows this effect to be mitigated by insertion of short dark exposures between the Flat Frames as they are gathered.  Also, if you take enough light frames to allow meaningful outlier pixel rejection during integration, then the Master Flat Frame can still be produced with little effect from the 10% of frames with a partial residual image in part of the frame in case you cannot control your light source from within SGP.

 

   I am still doing some more testing of Flat Frames and Bias / Short Dark Frames but have not been able to reach any new conclusions yet. There is still something else going on inside the sensor that I am trying to get a handle on. I'll post more, probably here in this thread, when I can make sense of it all.

 

 

John


Edited by jdupton, 25 June 2019 - 12:27 PM.


#35 PeteM

PeteM

    Vanguard

  • -----
  • Posts: 2218
  • Joined: 12 Jul 2009
  • Loc: West Michigan

Posted 25 June 2019 - 01:33 PM

John - Thanks for answering my question and confirming what I thought. I did gloss over that tread you mentioned but will go back a read it fully. Last night I posted a question here about a recent issue with a stacked image where the master flat did not seem to be correcting fully for the dust donuts. Almost like a ghost or residue was left. I will recapture my flats tonight with the technique above and see what happens. Guessing that since I did not follow the same protocol with the lights I might still have the issue. Luckily I have some clear skies on tap for the week, so I will give it try. If you need more additional testing from another camera, let me know. I have many cloudy nights in Michigan.



#36 jdupton

jdupton

    Surveyor 1

  • *****
  • topic starter
  • Posts: 1780
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 25 June 2019 - 02:27 PM

Pete,

 

   I don't think this effect will cause an incomplete correction of dust mote shadows in calibration but if your camera responds more like Andy's QHY294, then it might be possible. I would more suspect an incomplete calibration of the Master Flat itself for what you describe. If the Flat-Dark used to calibrate the raw Flat Frames before integration is off a little, then the flat could still leave faint versions of the dust mote donuts in your lights.

 

   One way to see if this is happening is to use one of your Flat Frames and subtract the Master Flat-Dark from it using PixelMath. In that subtraction, try adding or subtracting a very small offset like 0.0001 and then use that result as a trial Master Flat to calibrate one of your lights. Stretch the heck out of it to see if any residual dust motes show. You can then repeat with an adjustment to the 0.0001 value you added or subtracted to see if some value results in better calibration of the dust motes. Once you find a value that works better, do that adjustment to your real Master Flat and rerun a trial of calibrating your lights to see if it improves.

 

   I have done the above with my camera and found that my Master Flat Frame is sometimes off just enough that poor flat calibration of the light frames is a result. I haven't pinned down the exact conditions for it to happen but it appears to tell me that sometimes the Flat-Dark used to calibrate my Flat Frames is not a perfect match for the Flat Frames themselves.

 

   Back to your testing, you might also try to run an sequence like the one I ran in the RBI-Like thread (Post #52) to see if you sometimes get a partial residual image similar to what I see of if you can sometimes get a full frame residual ghost image like Andy saw with his QHY camera.

 

 

John




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics