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

Canon t3i/600D camera test results (read noise, gain, DR, well depth)

dslr imaging equipment astrophotography
  • Please log in to reply
14 replies to this topic

#1 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 19 September 2019 - 01:56 AM

I spent considerable time testing the properties of my Canon T3i/600D and would like to share my results with the community. I was motivated in doing so by 3 things:

  • Correcting what appears to be a. errors in this frequently cited post, and b. a lack of clear data for this camera esp. since sensorgen is no longer available. (e.g. imagers are often looking for the gain to use in PixInsight's Subframe selector process and are presumably using incorrect nos: ex1, ex2)
  • Understanding the behavior of my own specific camera
  • Developing a better understanding of the various terminology and testing procedures.

I'll describe my testing process in brief right after the results:

 

V3 Canon T3i test results
 

I largely followed the testing procedure shared by Craig Stark here and this description of CCD testing from the Handbook of Astronomical Image Processing. Testing makes use of the following statistical properties:

  1. The variance of a diff of images (bias, flats) is an estimate of twice the variance in one of the individual images (and addresses variations in individual pixel means, assuming the noise is the same across pixels)
  2. When taking flats, fluctuations in photons striking sensor and subsequent electron charge in the pixel follow Poisson statistics (Mean = Variance) 

 

  • The Gain and Read noise were each measured 4 times with diff data image pairs at each ISO
  • All image analysis was performed in PixInsight
  • When taking image diffs for variance measurement, a constant was added to ensure no clipping of values
  • Since I'm dealing with an OSC, when measuring diff of flats, I only used nos. from one channel (Red) 
  • To ensure vignetting in flats did not affect results, I used a crop of the central area for flats for mean and variance eval
  • I did not bother with dark current measurements for two reasons: 1. it is heavily temperature dependent and there is no way to truly measure or control the sensor temperature; 2. estimation will have to be indirect (through noise) since Canon does some auto dark current compensation for every image

 

Some notable observations:

  1. The Mean Bias signal at all ISOs is ~2048 DN
  2. Max ADU at saturation at all ISOs except ISO 100 is ~15831. At ISO 100 it is 14043. This means that of the 16383 DN (14 bit), only 15831 - 2048 = 13783 are actually used. Less at ISO 100
  3. Full Well capacity based on ISO 100 data is 24828 electrons
  4. Unity Gain is roughly at ISO 200. All gains are significantly different from Alexander Vasinin's work here which is the result that shows up most often during a search for gain data for this camera.
  5. The results reinforce my decision to image at ISO 800. It roughly marks the start of the ISO less state of the camera (read noise roughly varies linearly with ISO above this). Any lower and the read noise is much higher. Any higher and I'm losing dynamic range and amplifying other noise without any benefit.

Questions, feedback welcome. Hope this is useful for other T3i users.


Edited by arvindsh, 19 September 2019 - 05:08 PM.

  • SandyHouTex likes this

#2 sg6

sg6

    Fly Me to the Moon

  • *****
  • Posts: 5902
  • Joined: 14 Feb 2010
  • Loc: Norfolk, UK.

Posted 19 September 2019 - 02:53 AM

Unsure of the results and their implication but the Willbell reference refers it would seem to CCD, the Canon 600d is a CMOS and different technology. Well depths and noise alone are different between to 2 technologies.

 

Are references to CCD measurements and processes applicable?

I see/read a lot of instances of people "buying a CCD" when the reality is they are buying a CMOS.



#3 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 19 September 2019 - 05:00 PM

Yes, the terminology (read noise, gain et.) are the same for both sensor types and the same evaluation methods apply. As mentioned in my post, dark current is the one that poses a challenge for DSLRs since these cameras auto subtract some dark current estimates even in the 'RAW' image.



#4 BQ Octantis

BQ Octantis

    Surveyor 1

  • *****
  • Posts: 1590
  • Joined: 29 Apr 2017
  • Loc: Red Centre, NT, Australia

Posted 20 September 2019 - 07:15 AM

Thanks for the data!

 

I had seen Vasenin's work…in spite of the differences between your analyses, your conclusion of ISO800 is the same as his.

 

I get too much walking noise in my DSO shots at ISO800. It's probably technique…I do everything wrong. So I like using ISO1600—it gives me better results. ISO3200 gives me too much glow around the edges. For planetary, ISO2400 has been my go-to—through my f/15 Mak-Cass, it's the sweet spot between sharpness (from depth of field benefits) and SNR. But I do that all wrong, too…

 

Cheers,

 

BQ

T3i user


Edited by BQ Octantis, 20 September 2019 - 07:16 AM.

  • arvindsh likes this

#5 sharkmelley

sharkmelley

    Vanguard

  • *****
  • Posts: 2316
  • Joined: 19 Feb 2013

Posted 20 September 2019 - 01:43 PM

Well done!  I have a 600D and the results you have obtained are very similar to my own.  As you have discovered, unity gain is definitely near ISO 200 and not ISO 800, as misreported by Alexander.

 

You might be interested in a PixInsight script I wrote, latest version is here:

https://www.cloudyni...tics/?p=6677614

 

Mark


Edited by sharkmelley, 20 September 2019 - 03:43 PM.

  • arvindsh likes this

#6 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 20 September 2019 - 05:11 PM

Thanks for the data!

 

I had seen Vasenin's work…in spite of the differences between your analyses, your conclusion of ISO800 is the same as his.

 

I get too much walking noise in my DSO shots at ISO800. It's probably technique…I do everything wrong. So I like using ISO1600—it gives me better results. ISO3200 gives me too much glow around the edges. For planetary, ISO2400 has been my go-to—through my f/15 Mak-Cass, it's the sweet spot between sharpness (from depth of field benefits) and SNR. But I do that all wrong, too…

 

Cheers,

 

BQ

T3i user

 

Hi BQ, you're right that we reach the same conclusion, but for very different reasons. While what matters is having settings we are comfortable using, obtaining accurate parameters is important to develop understanding and model camera performance. 

 

Are you sure it's walking noise? I'd imagine it should show up at ISO1600 as much as 800.

 

 

Well done!  I have a 600D and the results you have obtained are very similar to my own.  As you have discovered, unity gain is definitely near ISO 200 and not ISO 800, as misreported by Alexander.

 

You might be interested in a PixInsight script I wrote, latest version is here:

https://www.cloudyni...tics/?p=6677614

 

Mark

 

Thank you, Mark. I've made a note to give your script a go.


Edited by arvindsh, 20 September 2019 - 05:49 PM.


#7 BQ Octantis

BQ Octantis

    Surveyor 1

  • *****
  • Posts: 1590
  • Joined: 29 Apr 2017
  • Loc: Red Centre, NT, Australia

Posted 20 September 2019 - 06:26 PM

Are you sure it's walking noise? I'd imagine it should show up at ISO1600 as much as 800.

Nope, not sure at all…again, I totally suspect it's my processing chain. But I remember a thread fusing DSLR sensor performance parameters and integration times (for the life of me, I can't find it). The conclusion was that ISO800 was ideal, but integration times needed to be over 60 seconds. Which brings me to the more practical aspect of ISO1600 and ISO3200: my mount. My Orion SkyView Pro leaves much to be desired for tracking performance. The probability that I get some sort of shift goes up exponentially with time, and my current sweet spot with an intervalometer and PEC playback is about 45 seconds at 200mm. And exposures on my much more transportable EQ2 mount quickly go pear-shaped beyond 30 seconds. So unless I want to cart out my guidescope setup (with its own set of embuggerances), I just use ISO1600. It brings me much joy.

 

I'm curious as to what drives the performance differences at the higher ISOs. While the e- saturation count is obviously lower, my DSO targets rarely exceed this (unless I integrate for too long at say 8-15mm FL). So for me, the clipping is just on the CA bleed around bright stars. So as most sources state, it my results should be the same for a DSO beyond ISO800—besides clipping, I should get the same results with ISO1600 and ISO3200 with RAW. But ISO3200 gives me a terrible red ring around the periphery that I can't simply bias/dark/flat out, and it takes too much effort to clean it up. It looks like amp glow—indeed, it gets worse the longer I image. So all things being equal, what's driving the higher current? Signal amplification? DIGIC noise reduction algorithms? Does it go away if I switch from JPEG+RAW to just RAW? Is there some other setting burried somewhere to mitigate it?

 

BQ


  • arvindsh likes this

#8 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 23 September 2019 - 03:37 PM

For short exposures the higher ISO likely helps boost signal meaningfully over background—at the cost of some loss of dynamic range. 

Technically, past the ISOless range, one should still get about the same SNR with lower dynamic range. In practice, for me at least, 5 min exposures at ISO 1600 or higher, results in too much noise. I presume, as you suggest, this is because of the inbuilt Canon dark current subrtration that keeps the bias constant at ~2048DN at all ISO values.



#9 Alen K

Alen K

    Apollo

  • -----
  • Posts: 1077
  • Joined: 25 Nov 2009

Posted 23 September 2019 - 06:31 PM

Technically, past the ISOless range, one should still get about the same SNR with lower dynamic range. In practice, for me at least, 5 min exposures at ISO 1600 or higher, results in too much noise. 

I don't have a dog in this race but to me that statement right there suggests there is something about your camera's behavior that you do not fully understand (or have not fully characterized, if you prefer) despite all of your measurements. 



#10 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 23 September 2019 - 10:57 PM

I don't have a dog in this race but to me that statement right there suggests there is something about your camera's behavior that you do not fully understand (or have not fully characterized, if you prefer) despite all of your measurements. 

 

Always happy to be corrected and enhance my understanding if you're willing to help, Alen. Unfortunately just saying that I don't understand, doesn't by itself give me the means to correct shortcomings in my comprehension.

Cheers.


  • SandyHouTex likes this

#11 sharkmelley

sharkmelley

    Vanguard

  • *****
  • Posts: 2316
  • Joined: 19 Feb 2013

Posted 24 September 2019 - 12:15 AM

Always happy to be corrected and enhance my understanding if you're willing to help, Alen. Unfortunately just saying that I don't understand, doesn't by itself give me the means to correct shortcomings in my comprehension.

Cheers.

Your own data shows that sensor read noise continues on its downward path at ISO 1600 and ISO 3200.  But then you made a statement that implies the SNR starts to reduce at those ISOs.  Something doesn't add up.

 

Sure, ISO 800 might be a good compromise between read noise and dynamic range I don't immediately understand why the SNR should reduce at higher ISO.

 

Mark


  • arvindsh likes this

#12 arvindsh

arvindsh

    Lift Off

  • *****
  • topic starter
  • Posts: 22
  • Joined: 09 Jul 2016

Posted 24 September 2019 - 12:23 AM

Your own data shows that sensor read noise continues on its downward path at ISO 1600 and ISO 3200.  But then you made a statement that implies the SNR starts to reduce at those ISOs.  Something doesn't add up.

 

Sure, ISO 800 might be a good compromise between read noise and dynamic range I don't immediately understand why the SNR should reduce at higher ISO.

 

Mark

 

That's right, Mark. I do state that from experience, shooting at ISO 1600 results in more noise than ISO 800, but also acknowledge in my post, my inability to explain why that would be. Technically I only lose DR at higher ISOs but practical experience suggests other factors at play. I surmise (again in my post) that Canon's built in dark subtraction process that preserves bias at 2048 DN at all ISOs might have something to do with it. 


  • SandyHouTex likes this

#13 the Elf

the Elf

    Surveyor 1

  • *****
  • Posts: 1741
  • Joined: 06 Sep 2017
  • Loc: Germany

Posted 25 September 2019 - 02:04 PM

Thank you very much for the excellent work. ISO 800 is recommended almost everywhere and I have never used any other ISO. For me the most important information, on which I would base the purchase of a new camera is the amount of dark current as a function of temperature. I was wondering if the dark current is high compared to modern sensors. Alsas neiter for the 600D nor for any highly praised sensor anyone has data about dark current. Does anyone know a source for it? I have about 1000 dark frames of 2 cameras, all sorted to a dark library. I did not bother measuring anything, because if no data is available for new sensors how can I know they are better with respect to dark current?



#14 sharkmelley

sharkmelley

    Vanguard

  • *****
  • Posts: 2316
  • Joined: 19 Feb 2013

Posted 25 September 2019 - 02:33 PM

For me the most important information, on which I would base the purchase of a new camera is the amount of dark current as a function of temperature. I was wondering if the dark current is high compared to modern sensors. Alsas neiter for the 600D nor for any highly praised sensor anyone has data about dark current. Does anyone know a source for it?

The sensor self-heating of the canon 600D during a session of long exposures is possibly the worst I have ever measured:

 

DarkCurrentGraphs4.jpg

[click on image to see a better quality version]

 

Mark


Edited by sharkmelley, 25 September 2019 - 02:35 PM.


#15 the Elf

the Elf

    Surveyor 1

  • *****
  • Posts: 1741
  • Joined: 06 Sep 2017
  • Loc: Germany

Posted 26 September 2019 - 12:55 AM

Mark, 

 

thank you very much. It is more than I ever found, but it is only partly an answer to the question. Is the dark current so high because the sensor temperature is so high (poor cooling, high energy consumption, etc.) or is the sensor temperature on the same level as the others and dark current is high for the same temperature?

I guess you made the measurements with the lid on and indoors. What I see outside is quite different. With the camera attached to the scope and looking upwards free convection can transport heat from the sensor front into a larger volume that is given by the draw tube and the reducer in the draw tube. Most of it is aluminum, transporting heat well. I leave the display flipped away to allow heat from the camera rear to be taken away be the wind that we always have. When taking 15 min subs the first one is cooler, from that time on the temperature is constant. If(!) the exif data sensor temperature value is correct and if (what I doubt less than the former) my fridge thermometer on the same level as the camera at the tripod leg is correct I have a constant difference of about 6-8°C from ambient to sensor. When taking darks with the lid on in the fridge the behavior is as shown in your graphs. 

Do you have temperature readings to the data shown? 

 

"The sensor self-heating of the canon 600D during a session of long exposures is possibly the worst I have ever measured:" It depends on to what you compare. My GF got a EOS10D free some years ago and that thing had visible noise even in daytime image. I was stupid enough to order a mono modded 20D in the Netherlands. The noise level was awful. Compared to these I love my 600D/T3i for the low noise. I found your cooler installation on your website. Well done. I am wondering if I try to add coolers to my cameras or rather go for two new models and let Brent convert one again. I'm a bit afraid of the dew problem. I'd try to build a regulated and keep the temperature above the dew point. The regulation is not a big problem for me, I'm an electrical engineer and know how to design electric circuits.

 

the Elf




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





Also tagged with one or more of these keywords: dslr, imaging, equipment, astrophotography



Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics