36 replies to this topic

[jdupton]

Interested ASI294MC Camera Folks,

 

   OK, here I go again. This exceedingly long diatribe covers what I have learned about the ASI294MC over the past few months. Be forewarned. This write-up filled up 8 pages of text in OpenOffice Writer and will have to be split into multiple postings here.

 

TLDR / Executive Summary

-  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -

   With quite a bit of testing of this camera, I now believe that calibration issues some users have reported with this camera are due to not one but several factors. I think these may be overcome but some users will balk at or simply disagree with the methodologies suggested here to improve calibration. While this information is based on considerable indoor bench testing of the camera, limited time under the stars has been available for full corroboration.

-  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -  -

 

   I was not sure whether to put this post in one of the several ongoing ASI294MC threads or start a new topic specifically on ASI294MC Image calibration. I guess I will start here as a separate thread and post a couple of pointers to here as appropriate.

 

   There has been a lot written on the "problems" with calibrating images from the ASI294MC camera as well as “problems” with the camera itself. Much of what I read was short on data and long on opinions, often from folks who have never used the camera. I had long been interested in this camera as a complement to my SXVR-H694 mono camera. I wanted an OSC camera with similar characteristics in terms of pixel size and overall chip size and field of view. Back in the middle of July, I purchased the ASI294MC and have been besieged with cloudy, rainy weather ever since. I have only had three sessions under the stars so far. I hope my luck changes so that I can get back outside with the camera.

 

   I had researched the camera and read all I could from actual users of the camera. I have tried to contribute to the discussion about why this camera may behave differently from CCDs and other CMOS camera offerings. CMOS users have had to adapt new methods for calibration compared to the existing well established methodologies long used for CCD cameras. I suspect this camera could need some adaptation of our methods once again. I do think this camera still has a shot at being a very nice addition to the astrophotographer's tool bag. Here are some background links to a few of my previous posts.

 

Comments on Sensor Packaging Notes:

https://www.cloudyni...dpost&p=8635528

 

Comments on Sensor Cooling Notes:
https://www.cloudyni...dpost&p=8789647

 

Musings on Outlandish Camera Calibration Methods
https://www.cloudyni...dpost&p=8700610

 

   All of the boredom brought on by our miserable weather lead me to do a lot of characterization of the camera to try to understand why some folks have problems such as those discussed in various threads here on CloudyNights and elsewhere. I have come to believe that the issues are due to no one single problem but rather several idiosyncrasies that can combine to cause issues with the calibration of images. I have reached my conclusions after shooting more than 3,500 Bias, Dark, and Flat frames spanning exposures from 0 to 960 seconds. I have done extensive analysis of the data and have sliced and diced it in many different ways as I investigated what I was seeing.

 

   I will preface all this by saying that I still don't completely understand and cannot yet explain all of what I am seeing. The following overly long article will just relate what I am seeing as I run tests on my camera. My conclusions and (soft) assertions are based on my testing to date. Due to the volume of material, I will have to split this document into multiple posting on this thread.

 

   My observations lead me to believe that the issues folks encounter from this camera can stem from the following areas:

• A Bias isn't always a Bias (in the sense of that obtained from a CCD camera)
   
• Short exposures (of less than 2 to 3 seconds) should be avoided (including Flats and Flat-Darks)
   
• Stable thermal control of the BSI (Back Side Illuminated) sensor is critical
   
• Long exposures should be followed with short Bias-like exposures for best results
   
• Fixed Pattern Noise in Bias Frames is color channel dependent

 

   On the plus side, I have also concluded the following from my testing:

• The "amp glow" seen from the sensor is very linear with respect to exposure length
   
• Dark Frames may be successfully scaled with proper calibration

 

What Is A Bias Frame?

 

   In my opinion, a contributor to calibration difficulties with this camera is an unexpected behavior at short exposure times.

 

   In the CCD world, a Bias Frame is a dark exposure of zero seconds. It is used in calibration to eliminate the Fixed Pattern noise of the readout row(s) of the sensor as well as remove the Offset added to the A/D measurement result. Many to most CCD sensors can be made to produce this zero exposure time image frame and readout. On the other hand, many to most CMOS sensors are designed for the video market and have finite (but small) minimum exposure times. I suspect compromises may have been made in the design of the sensor and its operation with this single (video) market in mind.

 

   We often think of a Bias Frame as we would a Dark Frame. We (or least I) have always assumed that a Bias Frame is exactly equivalent to  a Dark Frame of 0 second exposure length. My testing of the ASI294MC has shown that this is not true for this camera. Any use of a Bias Frame straight out of the camera in calibration will result in an invalid calibration of the target frame. This is also true for other CMOS cameras and is one of the reasons calibration is done with uncalibrated Dark Frames and uncalibrated Flat-Dark Frames.

 

   While many may still be using Bias-like Frames for calibration of the ASI294MC, following graph should show why this is problematic. The plot is from a test run with my camera using exposures of 0, 0.2, 0.4, 0.6, 0.8, 1.0, 1.2, 1.4, 1.6, 1.8, 2.0, 3.0, 4.0, 5.0, and 6.0 seconds. A total of 140 frames were taken at 0 seconds and 10 frames each at all other exposure values. The Mean ADU value for each set of frames is plotted. The sensor was cooled to 0° C for these tests. Gain and Offset were set to Unity at g=120, o=30 in the ASCOM Driver. SGP (Sequence Generator Pro) was used for image frame capture.

 

   In the plot, note in particular that the value of the Mean ADU at an exposure time of 0 seconds is greater than the Mean value at all exposures greater than 1 second but less than 3 seconds. Recall that a Bias (at exposure of 0 seconds) should equal the Mean ADU value of a Dark frame at 0 seconds. If we extrapolate the slope of the Dark Frames greater than 3 or 4 seconds back to 0, we find that a CCD-like Bias should have a value of 1914.04 ADU rather than the value of 1914.36 that is measured. For this camera, using a Bias out of the camera from calibration will give us some problems down the line. While the problem here seems small, the error can become noticeable after calibration after we stack our image calibration and target frames and try to aggressively stretch an image in post processing.

 

Fig. 1 - Mean Frame ADU Value vs Exposure Time (@ 0° C, Gain = 120, Offset = 30)

 

Short Exposure Timing For A CMOS SOC Sensor

 

   In the plot above, also note the sharp increase in ADU values for frames between 0.2 and 0.8 seconds. I believe this is caused by timing these short exposures with circuitry residing on the sensor itself. Unlike most CCD camera sensors, CMOS camera sensors implement an SOC (System On a Chip) whereby many other imaging system functions are implemented right on the same silicon sensor chip. In the CCD world, these additional functions in the camera circuitry are external to the sensor itself.

 

   Most Astro-Camera designs use in-camera timing for short exposures and the PC Device Driver timing for longer exposures. From the plot above, I believe the SOC is being used for all exposures up to 1 second. The sharp slope of the plot for values less than 1 second is most likely due to extra heat generated by on-sensor timer circuitry being active during the exposure. Above one second, the PC running the camera takes over the exposure timing functions. The slight inconsistency of the ADU values between 1 and 2 (or 3) seconds is due, I believe, to inconsistent timing from the device driver. PC systems are generally not good at real time operations. Longer times are more accurate as a percentage of the measured value. That is why things smooth out to a linear slope by the time we get to 4 second and greater exposures.

 

   One side effect of this SOC timing of sub-second exposures is that amp glow becomes quite visible in the short exposures. If sub-second exposures are used with this camera (for Flats and Flat-Darks for instance), calibration issues may arise if everything is not matched exactly. As an example, looking at the plot above, think of what would happen if you calibrate a 1.5 second Flat with a 0.5 second "Bias" or Flat-Dark. (I think this practice may be common since most of us have come to think that Dark current in a 0.5 second exposure can be safely ignored.) If we do a calibration like this example, then the calibrated Flat will not be a true Flat and is likely to cause problems in the calibrated image later. Residual "negative amp glow" will be present and introduce gradients that are likely to be hard to account for later in post-processing.

 

   This phenomenon of increased amp glow and heat generation / Dark Current for short exposures is present at all Gain and Offset values I have tested. The plot below shows similar data for Gains of 200, 300, and 390 in addition to the Unity Gain (g=120) plot shown before. Note that the slopes of both short and longer exposures become steeper as would be expected from increasing gain values.

 

Fig 2 - Mean ADU Value vs Exposure Time (@ 0° C, Gains = 200, 300, 390; Offset = 30)

 

Thermal Control Stability Of A BSI Sensor

 

   The next area of difficulty with calibration of images from this camera is, I believe, associated with the nature of the BSI sensor construction. I think we will see similar effects as more large area BSI sensor cameras become available in the future.

 

   A short introduction to BSI sensor construction is in order. “Normal” sensor construction is shown below. Here, the light sensitive pixels reside on the top of the sensor chip along with the chip control and SOC circuitry. The contact points to connect to the chip are also on the top side of the chip. Electrical connections are made between the top of the chip and the top of the ceramic carrier using short wires bonded to each. The back of the chip is bonded directly to the ceramic chip carrier with a special adhesive. This gives a lot of area of thermal contact. The cold finger of the camera is in contact with the bottom of the ceramic carrier. Any heat generated by the chip has lots of thermal contact area down through the ceramic carrier to the cold finger. Changes in temperature of the sensor chip due to self-heating and changes in set-point control of the cold finger are quickly equalized because of the extensive thermal contact area between all parts.

 

Fig 3 - Diagram of a typical Wire Bonded Sensor Chip Construction

 

   Now, a BSI sensor is constructed differently as shown in the next figure below. In this case, the light sensitive pixels are at the top of the stack which is actually the back side of the silicon chip. The chip control and SOC circuitry reside on the bottom of the chip is this view. The electrical contacts for the chip also reside on the back / bottom side. Electrical connections are made between the silicon sensor chip and the ceramic carrier with the use of microscopic solder balls. Thermal contact between the chip and the carrier is solely through these solder balls. There are at least 248 solder balls making the connections. (There may be a few more to supply power in parallel to the chip. These extra connections are not brought out through the ceramic carrier. The carrier itself has a total of only 248 connections to the circuit board of the camera.) Solder balls in this type of application tend to be between 30 uM to 50 uM in size. Thus the total thermal contact area is much, much smaller than the thermal contact area for a non-BSI sensor chip. (The ratio of thermal contact area between a BSI and non-BSI sensor like the IMX294CJK is about 0.5 to 248 if my calculations are right.)

 

Fig 4 - Diagram of typical BSI Sensor Chip Construction

 

   In the case of BSI sensors, thermal conduction between the silicon chip and the cold finger under the ceramic carrier is limited by the limited number of solder balls in the path. For BSI sensors, this thermal “bottle-necking or throttling” has some implications. Under steady state conditions, temperature control is not much different from non-BSI sensors. Both types of packaging can reach similar temperatures – they just equalize at different rates. However, when extra heat is generated by the sensor, it will take longer to stabilize to the temperature of the ceramic carrier and cold finger. Further, changes to the set-point temperature of the cold finger will take longer to normalize at the sensor chip compared to non-BSI structures.

 

   In my mind, the above observations have implications to how we capture images with the camera. I think it is prudent to allow some additional “thermal soak time” after reaching the desired set-point temperature. Since the temperatures on the chip take longer to reach steady state after quick changes, the sensor probably won’t reached the same stable temperature that is measured at the ceramic carrier.

 

   In the testing of my camera, I noticed that the stability of the sensor / carrier temperature can vary under some conditions. I have found that for very deep cooling where the TEC is operating at high power loads, the variation of temperature in the images is less stable than when the TEC is operating at lower power levels. Because the silicon sensor chip lags behind changes at the ceramic carrier, that likely means that sensor temperatures will also not be completely stable. After a number of ad-hoc experiments, I have set a personal goal of not running the TEC at more than 75% to 80% of full power. This can be easily verified with any camera by running the TEC at various power levels (and set-points) and then examining the variation in actual temperature recorded in the FITs Header of the captured images. A plot of recorded temperature over time for an imaging run (even with Dark Frames) can be very telling. In addition, a slow change in median ADU values for Dark or Bias Frames can be seen when the data is plotted against time. Furthermore, I have noted inconsistencies in the Dark Frame ADU levels for several minutes after the camera reaches the desired set-point.

 

   At set-points that use the TEC to 50% power or less, the variation is held to +/-0.2° C for the most part in my camera. Mean values of the Dark Frames at a given temperature appear quite stable after 5 to 10 minutes into a run. At TEC power levels higher than 80%, variations of up to 1° C were noted – usually warmer. As TEC power approached 95%+, even more variations were seen and the overall recorded temperatures rose as time went by. I attribute this rise to self-heating of the camera body as a whole. The temperature of the body of my camera rose considerably at TEC Power levels above 95% when compared to 80% or less power. I think the camera’s fan reaches its limit for extracting heat and the excess TEC heat is then dumped into the camera body rather than the surrounding ambient air which will indirectly affect the sensor temperature after a period of time.

 

   It is my opinion after testing my camera that temperature stability is more important than absolute sensor cooling in terms of getting steady repeatable Dark Frame Mean ADU values. Cooling to -30° C or less below ambient gave better calibration results than cooling more deeply. Since Dark Current is rather strongly influenced by sensor temperature, this has an effect on calibration of image frames. As I live in an area with quite warm Summers, I tend not to cool more than about 20° C to 25° C below ambient. (I actually have to stretch that in mid-Summer when my ambient daytime temperatures approach and exceed 40° C. At such extremes, it is hard for me to even reach my modest goal of cooling to 0° C before midnight.) A cooling of 25° C below ambient generally puts me in the range of 40% to 60% TEC Power with good thermal stability through a session.

 

 

John

 

Part Two To Follow (Multiple Long Exposure Result Peculiarities)

Edited by jdupton, 10 October 2018 - 10:38 PM.

[jdupton]

Interested ASI294MC Camera Folks,

 

   This is Part Two of my camera testing notes.

 

Multiple Long Exposure Result Peculiarities

 

   This next observation is still a total mystery to me. Early on in my experiments with the camera, I noted that the Mean ADU values for Dark Frames tended to be different from one session to the next on different days. In other words, I could cool to 0° C one day, wait for temperatures to stabilize for five to ten minutes and then start gathering Dark Frames. The Mean ADU values obtained would sometimes (but not always) be slightly different than those gathered under the same conditions on another day. I had trouble pinning down with any accuracy exactly what the Dark Current measurement from the camera should be as it appeared to differ from one day to the next. These were indoor tests run at the same (+/- 2° C) ambient conditions with the same set-point temperature, camera gain, and offset.

 

   In looking over my results, I noticed that for sequences where I also gathered Bias Frames (zero exposure Dark Frames) or short 1 to 4 second Dark frames, the values were very consistent session to session. I tried a new experiment several times. I placed a short exposure between all the long Dark Frame exposures I was running. It makes no sense to me but the Dark Frame Mean ADU values were much more consistent when captured with such a sequence. For all testing after this observation, I ran short exposures interspersed in the sequences and seemed to get better results. If I wanted to gather a Dark Library of 60, 120, and 240 seconds, I created a sequence with events for 1, 60, 1, 120, 1, and 240 second exposures and had Sequence Generator Pro rotate through the events rather than finishing entire events in order. Doing this on multiple days and in multiple sessions gave good consistency of Mean ADU values in the Dark Frames.

 

   I have no idea what to make of this anomaly. It just seemed to work for me. I intend to go back and design some new experiments to investigate this effect further.

 

   The implications of this on calibration and imaging may be that light frames of a target could be affected in a similar manner to the Dark calibration frames. If we were to assume perfect skies during multiple imaging sessions, we would also expect very consistent Mean ADU values in the images. However, if this effect happens during an imaging project, the Mean ADU values after calibration could vary on a night to night basis affecting the final integration. Scaling of the images during the integration process may offset any problems caused by the inconsistencies but more testing for this effect would be a good idea.

 

Channel-Specific Fixed Pattern Noise in Bias Frames

 

   In threads dealing with calibration problems for this camera, I saw two comments / questions that I have recently dug into a bit.

 

Quote By Jon Rista in https://www.cloudyni...n/#entry8879975

One thing that will really stand out as a camera artifact is purple. Purple is a very unusual color in space. It is not necessarily unheard of...but usually it would be formed by mixing other colors, say distinctly blue reflection nebula with deeper pink hydrogen nebula. That is not a combination largely present in the outer field of Veil to my knowledge, and the kind of purple you have in there is particularly remniscent of dark current that has been demosaiced. I spent a lot of time trying to optimize the calibration of my 5D III, which had several bits of amp glow in the corners, as well as a bright amp glow to the side. Those glows were often the same purple after I had calibrated the color of the image in PI. Its a purple that is the hallmark of demosaiced camera junk.

 

Quote by Wes (User calypsob) in https://www.cloudyni...-5#entry8879502

Does the gradient appear in dark frames?

 

   So, these comments and question bring me to the following observation. The answer as to whether a color gradient shows up in my integrated Dark Frames is Yes, albeit indirectly. Below is an STF high stretch of the DeBayered integration of 50 Dark Frames of 120 seconds exposure each. We (or least I) rarely look at a DeBayered calibration frame but the question compelled me to do some investigating. Indeed, I do see a color gradient in my integrated Dark Frame. And, to top it off, it is rather purple in color! The purple appears most pronounced along the left side of the image.

 

Fig 5 - Raw Un-Calibrated DeBayered Dark Frame (@ 0° C, Gain = 120, Offset = 30)

 

   Above, I mentioned that the color gradient is an indirect contributor in the integrated Dark Frame. By that I mean that while it is present in the dark Frame, it does not appear to be caused by any differences in Dark Current between the RGB color channels. That can be demonstrated by the following image.

 

   Here we have a DeBayered Bias Frame integration from my camera. This is an integration of 50 exposures of 0 seconds each taken in the same session as the above 120 second Dark Frame. The gradient here is more pronounced because it is not diluted by the Mean Dark Current ADU value also present in the Dark Frame. This is just pure fixed pattern noise. Yet it clearly shows that the pattern is not fixed between the R, G, and B pixels. The clear gradient then begged the question: Is there a component of this gradient also present in the Dark Frame?

 

Fig 6 - Raw DeBayered Bias Frame (@ 0° C, Gain = 120, Offset = 30)

 

   The next image shown below attempts to answer that question. This is the same Dark Frame above calibrated with the Bias Frame shown above and then DeBayered. To my eye, the purple gradient is gone suggesting that there are no differential color gradients within the Dark Current portion of the Dark Frame and the Color Gradient that is present comes in the form of a fixed pattern noise contained solely within the Bias portion of frame’s signal. (Note that this calibration of the Dark Frame was not done directly by the Bias Frame because of my comments earlier that a Bias Frame from the camera is not the same as a Dark Frame of 0 second duration. More on this later.)

 

Fig 7 - Calibrated DeBayered Dark Frame (@ 0° C, Gain = 120, Offset = 30)

 

 

John

 

Part Three To Follow (Scaling Of Dark Frames Is Possible)

Edited by jdupton, 10 October 2018 - 10:55 PM.

[jdupton]

Interested ASI294MC Camera Folks,

 

   This is Part Three of my camera testing notes.

 

Scaling Of Dark Frames Is Possible

 

   Contrary to data presented from other CMOS cameras, I have found that Dark Frames can be successfully scaled with the ASI294MC camera. Special methods must be used if there is a desire to get well matched Dark Frames to light frames, Dark Frame scaling in PixInsight can be used. For many users, it may not be worth the trouble but in order to save an image for which you don’t matching Dark Frames, it could save the day.

 

   The convoluted process for scaling Dark Frames begins with having a Bias and Dark Frame library taken at the same Temperature, Gain, and Offset. The first step is to determine the Dark Current slope and Y intercept of a plot of multiple Dark Frame exposure times. My Dark Library uses 50 each 0, 60, 120, and 240 second exposures. The data for Mean ADU values of each average integrated frame exposure is plotted against Exposure time in a spreadsheet. The slope and Y intercept of such a graph is easily obtained using the LINEST() function. The results will give us the parameters we are looking for. The Slope gives us the Dark Current rate for the sensor while the Y intercept gives us the equivalent Mean ADU value for a CCD-like Bias Frame taken at 0 seconds exposure.

 

   This Dark Frame-derived Bias Mean just tells us what the mean of our camera Bias should have been. It is just a number and contains no information whatsoever about the pattern noise from our camera. Our actual Bias Frame from the camera will have the pattern noise we need but the Mean value is off what it should have been. The second step is to subtract the difference between the Mean of the Bias image and the Y intercept on a pixel by pixel basis. This can be done using PI PixelMath with an equation of “$T – (mean($T) – Y_Intercept_Of_Dark_Plot)”. After this adjustment, we now have an Adjusted Bias that can be used to calibrate a Dark Frame so that it can be scaled.

 

   Scaling the Dark Frame for use in calibrating our lights can best be done using PixelMath again. Here, we simply scale the Dark by the ratio of exposure times between what we have and what we need. For example if we have Dark Frames in our library for 60 and 120 seconds but took our target lights at 90 second exposures, we would use PixelMath on our 120 second Library Dark Frame and write “($T - Adjusted_Bias) * (90 / 120)”. This gives us the scaled Dark Frame for use in the ImageCalibration for our lights. We would also plug in the Adjusted Bias as the Bias file in ImageCalibration. A similar pre-calibration process should be used on the Flat Frame we will use for ImageCalibration.

 

   Here is the result of calibrating a 60 second Dark Frame using an Adjusted Bias and scaled 120 Second Dark Frame as described above. The FlatCountourPlot shows pretty clean removal of the Amp glow from the integrated Dark Frame. The result appeared to be excellent to me -- a very flat frame with only the Dark frame FPN and sample noise. The variation across the full frame showed no real hint of amp glows and was flat to about +/- 1e-. (For this test a small pedestal of 0.01 was added to the frame to prevent clipping.)

 

Fig 8 - 60 Second Dark Frame (@ 0° C, Gain = 120, Offset = 30)

 

Fig 9 - 60 Second Dark Frame Following Calibration with Scaled 120 Second Dark Frame (@ 0° C, Gain = 120, Offset = 30)

 

Fig 10 - FlatContourPlot Of 60 Second Dark Frame Calibrated with Scaled 120 second Dark Frame

 

Putting It All Together

 

   All of my testing has lead me to the following conclusions. I intend to implement these in my usage of the camera if I ever see a string of clear nighttime skies again.

• There are no color channel differences in the Dark Current from the camera
   
• There is a color channel gradient component in the Fixed Pattern Noise of a Bias Frame
   
• A Bias Frame directly from the camera should not be used as is for any calibration work
   
• A manual adjustment of Bias Frames must be performed if they are used for calibration work
   
• In general, frames at exposures between 0 and 3 seconds should not be used for calibration work
   
• The camera should be allowed to “thermally soak” for 5 to 10 minutes after reaching the desired set-point temperature
   
• Thermal stability of the cooling seems more important than extreme cooling. (It is best not to run the TEC at power levels greater than about 80%.)
   
• Frames of 1 second exposure duration taken between imaging frames helps with session to session consistency

   I do not have a definitive answer on how the purple and green gradients come about for some users of the camera while other users do not see them. I strongly suspect that it is slightly different from user to user. Some things come to mind as possible major contributing factors. The greatest contribution may come from using Flat Frame and / or Flat-Dark Frame exposures of less than 3 seconds as this could be a major contributor. Similarly, calibrating a Flat Frame with a pseudo-Bias frame of 0.1 to 0.9 second exposure (under the belief that Dark Current will be zero) will definitely lead to improper Flat calibration. If the Flat Frame calibration is off by just a little, the residual colored gradients in the Bias component will be imparted to the target image. They may be very small and go unnoticed until final post processing when the image is stretched to the max to get every drop of faint structure to show up.

 

   Another major contributor may be due to poor thermal stability resulting from running the cooling of the chip to the maximum possible. Above 90% TEC power levels, the thermal control of the chip appears to suffer in my testing. Add to that the “high thermal resistance” between the sensor and the cold finger caused by the BSI flip-chip solder ball attachment and Dark Current may be more variable than is desirable at the sensor. Variable Dark Current in long exposures will directly result in less effective calibration.

 

   I welcome others to duplicate my experiments and show their own results. I don’t think my camera is unique in any special way. I have only seen the gradients in my one 3 hour long exposure and then only after going out of my way to stretch the image more than I should have. Even in that one case, I did not notice the purplish streaks until I was very aggressive with the DynamicBackgroundExtraction process. Interestingly, the purplish streaks vaguely looked like Flat Frame contours. I need to clean up my flats better and make another run at calibration of the data.

 

   I have some really wild ideas of additional things to try. So far, I not seen anything that says images from the camera simply cannot be calibrated. I am approaching this as a challenge to find better ways to work with the camera. I still strongly believe that calibration methods may have to change slightly for BSI sensor cameras and look forward to newer larger such cameras in the future.

 

 

John

Edited by jdupton, 11 October 2018 - 09:21 AM.

[Zebenelgenubi]

John

 

Thank you for posting your measurements and analysis. I have owned my ZWOASI294mc_Pro for six months and had never seem any of the image artifacts that couldn't be calibrated out with standard processing techniques.  Seems that I have been following most of the items in you suggested list of camera  operating conditions without every seeing your list.  Thanks again of the effort you put forth in producing a very insightful and useful posting.

[freestar8n]

Thanks for the write up.  Some things don't surprise me and suggest things I would do anyway, while other things just emphasize these cameras are black boxes made by humans and may not work as expected.

 

I'm not surprised you need to stabilize the temps for a while to get them to equilibrate well.  I would do that anyway and it corresponds to the overall thermal mass equilibrating.

 

I'm not surprised there is weird stuff going on in the bias below 1-2 seconds or so.  That has been noticed by several people.

 

I can't explain the need to insert short exposures to get consistent results - but again it doesn't surprise me too much.  When I did a recent study of a cmos camera and pattern noise, I randomized the sequence of all exposures so there was no sequence dependence.

 

I'm not surprised dark scaling works pretty well - that's consistent with what I have seen. But I would still avoid it if possible, since it isn't too hard to match exposures in lights and darks.

 

As for the color gradients in bias - I'm not sure what that means since you are deBayering in order to see the color.  As long as those colors are consistent - they would never show after calibration.  Anything static that you see in the dark or bias is perfectly fine as long as it is consistent - whether it is before or after debayering.

 

So - the only thing of concern to me is the need to insert weird exposures during a long exposure sequence.  If that is true I agree it is worth investigating and possibly implementing.

 

Frank

[Astro_BC]

Thank you for putting in the effort in testing and documenting. As a casual observer in the astro cam space looking to make an informed purchase, I appreciate your scientific approach to the concerns surrounding this camera option.

[Mert]

Interesting write-up for sure, you've put in a lot of time!!

Would it be an idea to registrate the version of the 

camera driver software?

I have seen extreme changes between software versions!

Edited by Mert, 12 October 2018 - 10:56 AM.

[jdupton]

Zebenelgenubi,

 

John

 

Thank you for posting your measurements and analysis. I have owned my ZWOASI294mc_Pro for six months and had never seem any of the image artifacts that couldn't be calibrated out with standard processing techniques.  Seems that I have been following most of the items in you suggested list of camera  operating conditions without every seeing your list.  Thanks again of the effort you put forth in producing a very insightful and useful posting.

 

   Thank you for the comments.

 

   I think most of what is on my personal list of "best practices" for getting the most out of this camera are pretty much already followed by most imagers as common sense. There is not much really new there. I think the most important items are 1) not using short exposures because of non-linear sensor response, and 2) backing off a bit on cooling to the max to allow for better thermal stability because of the BSI construction of the sensor.

 

   Not everyone will gain from adopting all suggestions but most practices will not cause any new problems, either.

 

 

John

[wargrafix]

What constitutes short exposure? 60 seconds?

[jdupton]

Mert,

 

Interesting write-up for sure, you've put in a lot of time!!

Would it be an idea to registrate the version of the 

camera driver software?

I have seen extreme changes between software versions!

 

   Not only would it be a good idea, I completely overlooked adding the version information as I should have. 

 

   I have been using the same versions that were current when I got the camera in mid-July. They are:

• Native Driver Version -- 3.0.0.4
   
• ASCOM Driver Version -- 1.0.3.18
   
• SGP Version -- 2.6.1.106

   Both of the ASI camera drivers have been updated since July but I have not updated drivers on my system yet. Since I am still using the older version of SGP, I use only the ASCOM driver since the native driver support was added in the 3.X updates.

 

 

John

[jdupton]

wargrafix,

 

What constitutes short exposure? 60 seconds?

 

   No, nothing nearly that long.  In the context of my testing, a short exposure is anything less than 4 seconds. Since the response of the sensor is non-linear and a bit variable below about 3 seconds, 4 seconds or longer should lead to fewer calibration issues.

 

 

John

Edited by jdupton, 12 October 2018 - 05:50 PM.

[jdupton]

Frank,

 

   Thank you for the detail response and comments. There is some real food for thought in there.

 

Thanks for the write up.  Some things don't surprise me and suggest things I would do anyway, while other things just emphasize these cameras are black boxes made by humans and may not work as expected.

 

I'm not surprised you need to stabilize the temps for a while to get them to equilibrate well.  I would do that anyway and it corresponds to the overall thermal mass equilibrating.

 

I'm not surprised there is weird stuff going on in the bias below 1-2 seconds or so.  That has been noticed by several people.

 

I can't explain the need to insert short exposures to get consistent results - but again it doesn't surprise me too much.  When I did a recent study of a cmos camera and pattern noise, I randomized the sequence of all exposures so there was no sequence dependence.

 

I'm not surprised dark scaling works pretty well - that's consistent with what I have seen. But I would still avoid it if possible, since it isn't too hard to match exposures in lights and darks.

 

As for the color gradients in bias - I'm not sure what that means since you are deBayering in order to see the color.  As long as those colors are consistent - they would never show after calibration.  Anything static that you see in the dark or bias is perfectly fine as long as it is consistent - whether it is before or after debayering.

 

So - the only thing of concern to me is the need to insert weird exposures during a long exposure sequence.  If that is true I agree it is worth investigating and possibly implementing.

 

"I'm not surprised you need to stabilize the temps for a while to get them to equilibrate well.  I would do that anyway and it corresponds to the overall thermal mass equilibrating."

 

   Very true. It is always good advice to allow some extra time for thermal equalization with any sensor. The thing I wanted most to point out from my research is that construction differences for a BSI sensor are going to make it take even longer to equalize. Direct contact thermal conduction through the relatively small area occupied by the solder balls slows down the equalization process more than we might otherwise think.

 

"I'm not surprised there is weird stuff going on in the bias below 1-2 seconds or so.  That has been noticed by several people."

 

   Also very true. Jon Rista (I think) has posted similar curves of sensor response for very short exposures from the ASI1600MM camera. It also showed very odd response for these short exposures.

 

   For that matter, even my SXVR-H694 camera shows a discontinuity in the sensor response at 2 seconds. That marks the point at which timing is switched over from camera based timing to PC device driver based timing. 

 

   Since this probably happens in nearly all cameras, it is something every camera owner should verify if very short exposures are needed.

 

"I can't explain the need to insert short exposures to get consistent results - but again it doesn't surprise me too much.  When I did a recent study of a cmos camera and pattern noise, I randomized the sequence of all exposures so there was no sequence dependence."

 

"So - the only thing of concern to me is the need to insert weird exposures during a long exposure sequence.  If that is true I agree it is worth investigating and possibly implementing."

 

   I have a lot of trouble coming up with an explanation also. I somehow wonder if this may be a self-calibration mechanism for the sensor. I am thinking of something along the lines of the sensor needing some number of exposures in order to self-calibrate the dark current suppression circuitry on the chip. The hypothesis would be that after some number of exposures, the dark current suppression mechanism stabilizes and you get consistent reading from the A/Ds. Prior to that, the A/Ds references haven't been completely equalized giving slightly different conversions until the bias on the reference is established. That is total conjecture at this point. Maybe I can design an experiment to attempt to verify this suspicion.

 

   I do plan to try investigating this more as long as my poor weather continues. We are in for another five days of rain again. It seems to be never-ending.

 

"I'm not surprised dark scaling works pretty well - that's consistent with what I have seen. But I would still avoid it if possible, since it isn't too hard to match exposures in lights and darks."

 

   I agree. Scaling Dark Frames should be a last resort to save data. It is nice to know that for at least this camera, it can be done successfully.

 

"As for the color gradients in bias - I'm not sure what that means since you are deBayering in order to see the color.  As long as those colors are consistent - they would never show after calibration.  Anything static that you see in the dark or bias is perfectly fine as long as it is consistent - whether it is before or after debayering."

 

   I need to think about this some more. I was assuming that a poorly calibrated Flat might impart some of the color gradient onto your lights but maybe not. You comment about it having no effect so long as the colors are consistent has me questioning my thought process. What you say makes sense. I'll try some experiments to see if there are cases where the colored fixed pattern noise can make its way into a light during calibration.

 

   I did check on one aspect today. I used my existing data that had been taken at temperatures of -10, -5, 0, +5, and +10 degrees C. I wanted to see whether there was a temperature component to the colored gradients.

 

Edit 10/13 13:30 -- Disregard the comments and data from here to the next boundary for now. I will post a correction later.

--  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --

 

   There isn't.

 

   I tested this by DeBayering a stack of Bias Frames at +10° C and another stack from the same session taken at -10° C. I then subtracted one from the other adding a 0.01 pedestal to prevent clipping. I then examined the colored gradient in both the original files and the Difference files using PIs DynamicBackgroundExtraction process. The difference file was flat showing that there was no temperature dependence in the colored Fixed Pattern Noise. Also, the Background extracted from both +10 and -10 frames were identical. A sample is shown below. (I DeBayered these using SuperPixel to keep the colors pure.)

 

Fig 1 - Extracted Bias Background. Stack of 50 Bias Frames (High STF Stretch)

 

   I did the same thing with stacks of 960 second Dark Frames taken at +10 and -10 degrees C. Here I only had three samples to stack. As was the case with the Bias Frames, the difference was completely flat in terms of color. No color shift was detected between the Dark Frames at different temperatures. Even the extracted background of the calibrated Dark Frames looked surprisingly good and showed no significant color gradients. I concluded that any gradients in the Dark Frames are constant over temperature. Below is the extracted background for one of the 960 second Dark Frames. While at first glance the background looks a little odd, we have to remember that the amp glows are throwing off the stretch.

 

Fig 1 - Extracted Dark Background. Stack of 3 960 second Dark Frames (High STF Stretch)

 

Edit 10/13 13:30 -- End of data with errors.

--  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --  --

 

   I will be doing more testing as time permits and attempt to get a handle on the questions having to do with inconsistent results of exposure sequencing.

 

 

John

 

 

EDIT 10/13 13:30 -- Note that I have found an error in my extracted background images above. I inadvertently used an incorrect frame in my PI processing of the backgrounds. It will change some of the conclusions I made. I will update that portion of my response in a new posting either very late tonight or tomorrow.

Edited by jdupton, 13 October 2018 - 01:36 PM.

[andysea]

This is very interesting. I no longer own the 294 but I had empirically determined that I would get much better images with subs in the order of 2 minutes and with not too deep cooling. So my experience seems to somehow be substantiated by your experimentation.

[freestar8n]

 

   I will be doing more testing as time permits and attempt to get a handle on the questions having to do with inconsistent results of exposure sequencing.

 

 

John

Hi John-

Thanks for the reply - and thanks again for the detailed write up.

 

Regarding the color in the bias - I would take a single bias and subtract a master bias and see if that shows color after debayer.  You may need to add an offset somehow.

 

If you see color in the result - that would be an indication the color might actually show up after calibration.  But it may be happening at such a small level in terms of adu range that it may be negligible in practice - even if it shows in such a controlled test case.

 

Frank

[wargrafix]

You know whats funny about thread?

 

 

There is thought and methodical approach to understanding this issue. Clearly ZWO did none of those things. In my responses zwo give some halfway (thats being generous) explanations. And they have slowwwwwwww response times. like days between responses. Shows they are clueless about the issue. If it were not for budget I would have skipped ZWO this rounds.

 

Should I use ASCOM when selecting camera...or native camera in sharpcap?

[jgraham]

Hmmm, I had to send a camera back for repair and they did a great job. It spent longer in customs than it did in their shop. I thought that communications were fine considering that their work day is about as far outa kilter with mine as you can get. As for the ASI294MC Pro, there are other cameras out there. I thought QHY offered a camera based on the same chip. For me, this issue (that I still haven't noticed, but that's just me) is not a problem, but an operating characteristic that drops out during normal processing. As it is, I feel fortunate to have two great cameras at a price that I could afford. This would have been unthinkable just a few years ago.

What a great time to be an amateur astronomer.

[Jon Rista]

Frank,

 

   Thank you for the detail response and comments. There is some real food for thought in there.

 

 

"I'm not surprised you need to stabilize the temps for a while to get them to equilibrate well.  I would do that anyway and it corresponds to the overall thermal mass equilibrating."

 

   Very true. It is always good advice to allow some extra time for thermal equalization with any sensor. The thing I wanted most to point out from my research is that construction differences for a BSI sensor are going to make it take even longer to equalize. Direct contact thermal conduction through the relatively small area occupied by the solder balls slows down the equalization process more than we might otherwise think.

 

"I'm not surprised there is weird stuff going on in the bias below 1-2 seconds or so.  That has been noticed by several people."

 

   Also very true. Jon Rista (I think) has posted similar curves of sensor response for very short exposures from the ASI1600MM camera. It also showed very odd response for these short exposures.

 

   For that matter, even my SXVR-H694 camera shows a discontinuity in the sensor response at 2 seconds. That marks the point at which timing is switched over from camera based timing to PC device driver based timing. 

 

   Since this probably happens in nearly all cameras, it is something every camera owner should verify if very short exposures are needed.

 

I did indeed, and a couple other people have as well.

 

 

It is interesting that other cameras are exhibiting similar issues. I'm a little surprised about the SXVR-H694... I wonder what causes this...

[jdupton]

Jon,

 

I did indeed, and a couple other people have as well.

 

It is interesting that other cameras are exhibiting similar issues. I'm a little surprised about the SXVR-H694... I wonder what causes this...

 

   On my SXVR-H694, there was no change in slope at the 2 second mark but only a slight discontinuity in the DN vs Exposure plot. I am sure it was caused by slight timing differences between the frequency of the on-camera hardware timer and the in-PC device driver software timer. It was of no real consequence as long as I avoided taking exposures that might straddle the 2 second mark. (Shoot 0.0 to 1.5 or 2.5 to infinity but avoid the 1.5 to 2.5 exposure value.) By avoiding 2 second exposures, it never caused any problems in my images.

 

 

John

[phtnnz]

Thanks for the detailed analysis. Finally an explanation what's wrong with the ASI294MC biases. I'm currently doing flats at 0.3s and darkflats at the very same 0.3s exposure time with my ASI294MC Pro. If flats and flatdarks are exactly matched, this shouldn't be a problem, or however ... ?

 

Martin

[jdupton]

Martin,

 

   Welcome to CloudyNights!

 

   Yes, if you manage to get the Flats and Flat-Darks exactly matched, then there will be no problem. 

 

   The problem becomes how to make sure they are matched exactly. Gain and offset are easy -- temperature not so much. Note from the Bias and short Dark plot in Figure 1 of the first post, that the slope of the curve is much steeper at 0.3 seconds exposure than the slope of the curve at exposures greater than 4 seconds. This means that the average Dark Current is not only higher at 0.3 seconds but also since the slope is greater in that region, the chances of sensor self-heating can increase. With high slope, a relatively small difference in timing from exposure to exposure produces a larger difference in Mean ADU of the Frame.

 

   That may mean that beating on the sensor with repeated short Flats and/or Flat-Darks will probably degrade the temperature stability of the sensor during the run. My best guess is that temperature will change due to self-heating more during a run of multiple 0.3 second exposures than a run of 4+ second exposures. This can translate into having a harder time ensuring the actual sensor temperature matches between the Flat and Flat-Dark.

 

   If at all possible, I would try to dim the Flat source and run longer Flat Frame exposures. The alternative might be to insert a delay of 10 seconds or more between exposures to allow the temperatures more time to stabilize while shooting a long series of Flats and Flat-Darks.

 

 

John

Edited by jdupton, 13 October 2018 - 01:24 PM.

[phtnnz]

My flat field box is rather bright even at the lowest setting, I used unity gain 120. I’ll see what exposure times are possible with gain 0.

Martin


Gesendet von iPhone mit Tapatalk

[jdupton]

Martin,

 

   I would not go below the Unity Gain setting of 120. You can do that but you give up the low Read Noise characteristics of the sensor when you do. Read noise is just below 2 electrons at gain = 120 and jumps to about 7.5 electrons at a gain of 0. That will cost you quite a bit in terms of SNR (Signal to Noise Ratio).

 

   If it were me, I would get a sheet or two of Solar Darkening Film material such as can be used on car or home windows. It can be found at many auto supply stores and, here in the states, at most home improvement stores. It comes in various densities. You can try some of the %5 variety which passes about 5% of the total light hitting it. Just cover the flat panel and readjust the exposure to get back to the ADU you want to use.

 

 

John

Edited by jdupton, 13 October 2018 - 03:00 PM.

[phtnnz]

Thanks, John, for the hint. A quick search for darkening film found mostly 25-35 % transparency, I'll see what I can get.

Otherwise I thought about using gain 0 just for flats / flatdarks where SNR shouldn't be much of an issue?

 

Martin

[Jon Rista]

John, I wanted to thank you for taking my comments about purple gradients and examining your demosaiced darks and biases. I think that with OSC cameras, researching the dark signals in color is a very important part of fully understanding the characteristics of the dark signal. I have found it interesting that in most of my own work with DSLR/OSC, any time I find that particular purplish color in my images, I usually find that it correlates strongly with junk from the dark signal of the camera, and is not from space, and I am often then able to remove it with more careful pre-processing. 

 

 

Multiple Long Exposure Result Peculiarities

 

 

   The implications of this on calibration and imaging may be that light frames of a target could be affected in a similar manner to the Dark calibration frames. If we were to assume perfect skies during multiple imaging sessions, we would also expect very consistent Mean ADU values in the images. However, if this effect happens during an imaging project, the Mean ADU values after calibration could vary on a night to night basis affecting the final integration. Scaling of the images during the integration process may offset any problems caused by the inconsistencies but more testing for this effect would be a good idea.

 

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...

[jdupton]

Martin,

 

Thanks, John, for the hint. A quick search for darkening film found mostly 25-35 % transparency, I'll see what I can get.

Otherwise I thought about using gain 0 just for flats / flatdarks where SNR shouldn't be much of an issue?

 

   I think two layers of the 25% film would do the trick. Once you have some, you could always add another layer if required since a typical package has much more area than you are likely to need.

 

   Regarding, using Gain of 0 just for Flat and Flat-Dark frames is an interesting question. 

 

   You should start a new thread for that subject since it will apply to any camera and other users with different cameras might ask themselves the same question.

 

   On one level, my instinct says it might work. On another level my instinct says it may cause unforeseen problems. It is definitely a question worthy of a separate thread.

 

 

John

Edited by jdupton, 14 October 2018 - 08:16 PM.