Jump to content

  •  

* * * * *

Understanding the ZWO ASI 294MM Pro Camera


Discuss this article in our forums

Understanding the ZWO ASI 294MM Pro Camera

By Steven Bellavia

 

Like other new CMOS cameras being introduced into the astrophotography market, the ZWO ASI 294MM Pro seems to be a strange beast (but not in a bad way).

The first thing to notice is that it has a “jump” in performance at Gain 120.

I have highlighted the graph from the ZWO web page:

Chart, scatter chart

Description automatically generated

There is a drop in read noise from over 6 electrons to below 2 electrons, which causes a 2-stop increase in Dynamic Range, going from 11-stops to 13-stops.

 

But this is not where the strangeness ends.

That performance is in the “BIN2 mode”, which is supposed to be the default mode for the camera. (which it is not).

In BIN 1 mode, it becomes a 47 Mega-Pixel camera, with 2.31 micron pixels, performing as a 12-bit, high-resolution camera:

 

As you can see, no “jumps”, and 11 stop Dynamic Range at a similar Gain, with less than 2 electrons Read Noise.  Not bad for a high-resolution camera, and similar to the ZWO ASI 183, but with a larger format sensor.

 

So my problem:

How do I know if I have “activated” the 14-bit BIN2 mode, and I am not just 2x2 binning the 12-bit BIN 1 mode?

So I started by running a script in PixInsight called “Basic CCD Parameters”.

You take two identical Flat Frames, two identical Bias Frames, one 30-second Dark Frame and one 300-second Dark Frame, and enter them into the Script.  For all these images I selected 2x2 Bin from my image Capture software.  Below are the results:

Graphical user interface, table

Description automatically generated

Not good.  That is only 11 stops DR, and not the Full Well Count it should be, with a very strange Gain of 0.221    It should have been close to unity Gain (1.0 electrons/ ADU). But the read noise was low at 1.673 electrons. (An ADU is an Analog Digital Unit)

So was I not activating the BIN 2 mode, with the higher Dynamic Range?  How could I tell?

So I reverted to a measurement method that was developed either by Craig Stark or Richard Berry (I don’t know who came up with this first, but it is ingenious).  Nothing new.  From around 2003, and likely for CCD cameras, not CMOS, but it should work.

And why is it ingenious?  Here is the first part:

 

 

 

Part I -  Gain and Full Well Count

Light, i.e., photons, hitting pixels and emitting electrons (what Einstein got his only Nobel prize in) are discrete events, so they follow a Poisson Distribution.  Without getting too much into the mathematics, one very nice thing about the Poisson distribution is that the Standard Deviation = square root (Mean).  That is quite amazing in itself, and very fortunate for me.

So by doing several pairs of flat frames, and then plotting the Mean against the square of the Standard Deviation of the difference between each pair, you are essentially plotting the Mean against the Mean.  And anything plotted against itself should be a straight line with a slope of 1.0

But the ADC (Analog Digital Converter) in the camera is not reporting single electrons for each ADU it puts out.  That is a function of the Gain.  At low Gain, the camera is collecting several electrons for each ADU, and at high Gain, only a fraction of an electron for each ADU.  At Unity Gain, it becomes 1.0 electron/ADU. So the slope of the line of Mean versus the square of the standard deviation (the Variance) IS the Gain.  And then multiplying the Gain by the maximum ADU output, you also get the Full Well Count.  Genius!

So after converting all my measurement (done in AstroImageJ) to 14-bit (essentially just dividing all the numbers by 4), this does now look like I am getting the “correct” Gain, and Full Well Count.

Why did the PixInsight script show a different result?  I don’t know.

And because of that nagging question I had to go further.   So back to Craig Stark and Richard Berry.

 

Part II - Read Noise

By taking a number of Bias Frames (I did 60), and then subtracting out a single Bias Frame from the stack, you can determine the Standard Deviation.  This IS the Read Noise.  This is how much each Bias frame differs from the Mean.  (Noise, by definition, is the uncertainty.  Since this frame has no photons hitting the pixels, and is too short for dark current, the only remaining uncertainty is the Read Noise in the electronics chain).

So how am I going to use this to determine if I am getting the BIN 2 mode performance?

This was my idea (but standing on the shoulders of Stark and Berry):

Do Bias Frames in the BIN 1 mode, on either side of the “magical Gain” of 120, where the “jump” occurs.  Then repeat while binning 2x2, to see if there is the jump in the Standard Deviation, i.e., Read Noise.

And this is that result:

Graphical user interface, application, Word

Description automatically generated

And look at that!  When capturing images with a 1x1 Bin, there is no jump in Read Noise (StdDev), going from Gain 119 to Gain 121.  But when capturing images with a 2x2 Bin, the jump is very noticeable.

Also note that this is in 16-bit ADU’s.  It is less in electrons by at least a factor of 8 for the 12-bit BIN 1 mode, and at least factor of 4 less for the 14-bit BIN 2 mode. I say “at least” as you also need to multiply by the Gain, which might be a little less than 1.0 for BIN2 mode, based on the Gain test shown earlier.

Conclusion:

Simply by “asking” the camera to bin 2x2 turns the “switch” to allow the improved performance.

I later found out that this is in the latest firmware of the camera, so it is independent of the software asking it to bin 2x2.  It is not the “Default” mode, but very easy to get to with an image capture program. And I also just found out there is a BIN 3x3 mode, and a Bin 4x4 mode.  In 3x3 you are in the 12-bit mode and cut your resolution from the 47 mega-pixel 8288x5644 to 4144x2822.  Then if you go to bin 4x4 you get the performance boost, going from the 4144x2822 pixels, to 2072x1411.

Appendix A:  The ZWO ASI 294MM


  • Bob Campbell, dswtan, ERHAD and 2 others like this


7 Comments

How much of this might apply to the OSC version, i.e., ASI294MC PRO ?

    • Bob Campbell likes this

How much of this might apply to the OSC version, i.e., ASI294MC PRO ?

There is no corresponding bin1 mode in ASI294MC. The sensors are different, IMX294 in ASI294MC and IMX492 in ASI294MM. Not sure how much they are different besides CFA. The IMX294 is 'Quad Bayer' with similar 2.3um pixels grouped by 4 into larger pixels, then covered with CFA. Even if the sensor itself has similar bin1 mode, software would have hard time interpreting that funny Bayer pattern.

    • StevenBellavia likes this

Excellent article and I was following along, surprisingly quite well, until the end where I got lost. So can you kind of explain again in the binning for dummies version.

 

bin 1x1 at gain 120 is......

bin 2x2 at gain 120...  etc...  

 

THANK YOU

 

Dawn

    • StevenBellavia and jmfdiver like this
Great info I would that same type of article for the 2600MC pro
    • StevenBellavia likes this
Photo
StevenBellavia
Jun 28 2021 08:26 PM

Excellent article and I was following along, surprisingly quite well, until the end where I got lost. So can you kind of explain again in the binning for dummies version.

 

bin 1x1 at gain 120 is......

bin 2x2 at gain 120...  etc...  

 

THANK YOU

 

Dawn

Hi Dawn,

 

I am not sure I understand the question, because I am a dummy. 

But I will try to answer what I think you are asking:

 

In BIN 1 mode, there is nothing special about Gain 120, versus Gain 100, Or Gain 140, or even Gain 119 or Gain 121.  In BIN 1 mode, Gain 120 is close to unity gain

(A sensor analysis I performed in SharpCap shows it to be around 111 for my camera, which is oddly similar to the ZWO ASI 183, and that is pretty much how the 294MM performs in BIN 1 mode).  Gain 120 is just one of the many choices available*.

 

*The Gain you choose, also depends on many things, and I have developed spreadsheets that try to help narrow down the choice of gain and sub-exposure as a function of your sky's brightness, telescope and filter used:

 

https://www.cloudyni...or-zwo-cameras/

 

But in BIN 2 mode, Gain 120 is very special.  Yes, like BIN 1, it too is close to unity gain.  But a large increase in performance occurs at or above Gain 120 (and I don't know how they do it, as I am more of a math person than an electronics guru). Below Gain 120 there is 3X as much read noise, and you lose dynamic range, since that is a function of the full well count (which has a more gradual, steady change) and the read noise (which goes up dramatically below Gain 120).  And in fact, if you look at the BIN2 performance curve (attached /below), you don't catch up to the Gain 120 dynamic range until you get to Gain 0.

 

I don't like giving advice (as I am in no position to), but I don't see much benefit using Gains lower than 120 in BIN 2 mode. (but many people do, and I am sure with good reasons.  One reason might be just to be able to use longer sub-exposures.  And that is a good reason if you don't feel like filling up your hard drive. And many short sub-exposures are less efficient, due to download time between each exposure as well as dithering, etc., so that is another good reason).

 

So I hope that answers your question regarding  BIN 1 versus BIN 2 at Gain 120. 

 

I also hope I explained correctly in my article that going from BIN 1 to BIN 2 is simply achieved by changing the BIN setting in whatever image capture software you use, as it should have that option.

 

And lastly, whether you choose BIN 1 or BIN 2 opens up an entirely new discussion regarding sampling and resolution.  This then involves telescope aperture, f/#, atmospheric seeing conditions, guiding performance, final integrated SNR, as well as your personal preference in how your image "looks".  There are numerous threads on this topic. 

The table below shows what size pixel you should use to be "well-sampled", in terms of theoretical resolution, which is a function of f/#, (though atmospheric seeing has a big role in the final choice).  If you notice, that for any aperture, the f/# and pixel size determine if you are below, at or exceeding what the aperture of your scope can theoretically resolve.  So very briefly and simply:

 

f/#    Pixel size for matched theoretical resolution

4      2.40 microns

6      3.75

7      4.30

8      4.80

 

The reason you find most cameras with these common pixel sizes is that sensor manufacturers are not necessarily designing for astronomy, but all photography. That is why those sizes were chosen, since those are standard f/#'s common in photography.  In fact, if you look at the specs for a Canon sensor, for example, it will tell you the "Diffraction Limited Aperture" (DLA) that goes with that pixel size. (also below, which comes close to my calculations, though not an exact match).

 

I think I may have answered too much...    :/

 

Steve


Attached Thumbnails

  • Attached Image: ZWO_ASI294MM_BIN_2_Curves.png
  • Attached Image: diffraction_versus_sampling.png
  • Attached Image: Canon_Sensor_specs_EOS_60D.jpg
    • lowry_pt likes this

wow.  YOU more than gave an answer.  THANK YOU..  so essentially.  for BIN 1 ~~~Gain is a matter of desired exposure length. 

 

and then BIN 2.~~~ 120 or higher.  I am currently running Gain 200 for both actually~ ( another reading I found stated above 190 some lines or something). 

 

I really appreciate the explanation.  I think I wrapped my head around most of it.  I will reread it again.

 

Dawn

    • StevenBellavia likes this
Photo
StevenBellavia
Jun 29 2021 08:39 AM

wow.  YOU more than gave an answer.  THANK YOU..  so essentially.  for BIN 1 ~~~Gain is a matter of desired exposure length. 

 

and then BIN 2.~~~ 120 or higher.  I am currently running Gain 200 for both actually~ ( another reading I found stated above 190 some lines or something). 

 

I really appreciate the explanation.  I think I wrapped my head around most of it.  I will reread it again.

 

Dawn

Yes, the only "purpose" for a higher gain is to be able to do a shorter exposure.  This is no different from "regular" photography, such as being able to capture a ball deforming a tennis racquet, for example, with a very short exposure. 

But the total integration time required is essentially the same for all gains and exposures.

 

Gain and exposure are intimately tied to each other, and the best value for each "pair" should come from an optimization of SNR and Dynamic Range.  It is not only dependent on sky brightness, and f /#, but could be object dependent too.  Some objects need more attention to dynamic range (The Orion nebula, Andromeda galaxy), while others might need much more attention to SNR (Spaghetti nebula, Jones-1).

 

And of course, the highest priority needs to go to overall image quality.  There is no equation for that, and this is what makes this hobby so fascinating.

 

(:

 

Steve



Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics