Camera update from EOS 600D to EOS 800D (T3i to T7i)

At the time of writing in December 2019 I was using an unmodified Canon EOS 600D (T3i in the US) for my daytime photography and for astro photography as well. It was a purchase that dates back to May 2013 and at that time it was not brand new. Not too long after that so called ISO invariant cameras came up. First Nikons, Canon followed later. I started wondering if it is the right time to update to a more recent model. There are different opinions if Canon or Nikon is better suited for astro photography. Both have pros and cons. From what I watched on YouTube and found in articles Canon came up with the 800D and the 77D at the same time. They are technically quite the same but the 77D comes with the shoulder display. Later the 200D (SL2 in the US) came up with the same sensor as well its successor 250D (SL3). There are differences in auto focus sensors and video features and the 800D is the only one with the 2.5mm plug for remote trigger while all others come with the N3 connector known from old models. As I am used to the buttons positions and menus of my 600D pretty well I decided to go for the 800D which is very similar.
I was seriously thinking about getting a Nikon 5300, 5500 or 5600 because they have very good dynamic range values and they are ISO invariant, sometimes also called ISO-less which is technically incorrect. While the performance was outstanding for years now Canon caught up and offers almost the same performance. With quite some technical background from work I decided to measure the sensors of my old 600D and the new 800D. My girlfriend owns the 70D and the ancient 10D from 2003. Why not add them to test?

Sensor testing

For testing I followed the instructions in this book:
Digital SLR Astrophotography, Second Edition by Michael A. Covington, ISBN 978-1316639931
Mr. Convington made up loose groups of first, second and third generation cameras. The 10D is a member of the first generation that spoils the image with plenty of noise even at daytime. 600D and 70D belong to the second generation that is much better but obviously suffers from a lot of noise downstream, that is in or near the actual analog digital conversion. These camera are operated at higher ISO settings like 800 or 1600 to flood the downstream noise by plenty of signal. This of course cuts down the dynamic range. The 800D belongs to the third generation of (almost) ISO invariant cameras. At daytime this means if you shoot raw and your image is far too dark you can pull it up in some software and it will be as good as an image with a higher ISO setting. In an ideal world one could use them with the lowest natural ISO setting, probably 100 but as we will soon see, this is not quite the case.

How to measure

The aim of the game is to make up a chart with dynamic range on the y-axis and the ISO setting on the x-axis and a second chart that shows the read noise in electrons for each ISO setting. For a detailed explanation please refer to the book. Basically you need a software that can read out the actual digital numbers in an image and calculate the standard deviation. I used PixInsight. For each ISO setting two bias shots and an image with a saturated part is needed. With that little information the dynamic range and the read noise in digital number (DN) can be calculated. To convert read noise in DN to read noise in electrons the gain factor is calculated from two well exposed but not saturated flats. This is done at one medium ISO setting only because the conversion strictly doubles when going up one full ISO setting or is cut in half when going to a lower setting. I did that at ISO 400. Though not necessary for the bias and the saturated image I used the green pixels only. The test area was a 1000x1000 pixel area in the center of the sensor. Let's have a look at the dynamic range first:
The 10D is using a 12 bit ADC which is one reason for the low performance. The dynamic range at ISO 100 seems to be high but I doubt this is true. I rather think the coarse digital conversion leads to a too low estimation of the standard deviation. The saturation level is in the 4000ADU range, the bias is near 128ADU. The other three cameras use a 14 bit ADC. The members of the second generation 600D and 77D have a saturation value near 13500 at ISO 100 and over 15000 for all other ISO settings. Bias is near 2048. The third generation 800D behaves different. The saturation value is exactly the same for all ISO settings, including ISO 100 and it is 16383, exactly the maximum number of a 14 bit range. A second surprise is in the bias. While it is near 2048 for ISO 400 and higher it is cut down to 512 for ISO 100 and 200. This may be the reason for the slightly lower slope in the upper part of the curve. The best ISO setting for astro photography is where the curve turns to it's maximum slope which is ISO 800 for the 600D and the 77D. Although I do not recommend getting a 10D for astro photography today, if you do, ISO 200 is best.
Let's have a look at the read noise:
As expected the members of 1st and 2nd generation need higher ISO settings to get the read noise down. It levels out at the highest settings. The 3rd generation 800D behaves different. For ISO 100 and 200 the curve has a steep slope while it is almost flat for all other settings. Remembering the different bias values for ISO 100 and 200 this is not a big surprise. My conclusion is that I return the 70D and the 10D to my girlfriend. All jokes aside, of course I do and the main reason for doing all this was to find out if the 800D is an improvement and I think it is. Second reason was to know which ISO is best for the 800D (and probably all models with the same sensor). Both, the dynamic range and the read noise curve, tell me to use ISO 400 and this is what I did.
Before we move on, I would like to thank Michael Covington for his help and feedback on my data.

Real life test

To my great surprise the day the camera came in was followed by a clear night. It was pretty windy but clear, conditions good enough for a test. I decided to go for a shorter focal length and a bright target. So I pointed my 65mm quadruplet to the Seven Sisters or Pleiades. For a true comparison I decided to operate both cameras at he best ISO setting, 400 for the 800D(T7i) and 800 for the 600D(T3i). To compensate the difference in gain and use the higher dynamic range I took 4 minutes subs with the 800D and 2 minutes subs with the 600D. Both cameras supposed to collect the same over all integration time. It would have been nice and easy to image 2 hours with one and 2 hours with the other. As we all know that would not work well because the sky background the the transparency changes over night. So I did it the hard way. First of all I checked if perfect focus with one camera is still perfect focus for the other one. As expected this was the case. So I started with 4 subs of 4 minutes at ISO 400 with the 800D. That sums up to 16 minutes and a bit for dither. Unplugged the trigger cable, carefully removed the camera and replaced it with the 600D. Trigger in, exposure time 2 minutes, 8 subs, which is 16 minutes again. You know how the story goes on. This was repeated 5 times. Alas I needed to do a meridian flip during the session. Replacing a camera sometimes kicked the guide star out of the search field. As you for sure have guessed my focuser is not that rock solid and focus did not remain perfect during the session. The ever stronger wind added to the mediocrity of the images. PixInsights SubframeSelectoScript showed a poor 4 to 6 arcsec FWHM. I sorted out the worst subs and ended up with 19 x 4 minutes and 38 times 2 minutes. For the 600D I have a master bias and a dark library. To save some time and make the procedure even more complicated I took dark frames for the new 800D while her older sister was at work. Now you might say the test is in favor to the 600D because she was allowed to cool down while the 800D was at work all the time. True, but wait...

Processing

Thanks to the high speed of 5 images per second taking the 200 bias frames did not take long. Despite my memory upgrade to 16GB the stacking of the bias frames took very long. The darks were stacked as well an the result consists of 17 x 4 minutes. I took 30 flats for each one. Each camera's lights were calibrated using bias, f lats and darks and debayered in PixInsight. There was no cosmetic correction applied. Although one might take a different route I decided to star align all subs to the best one of the 600D, in other words the 24MPix images of the 800D have been scaled down to 18MPix. Of course this reduces noise. On the other hand the larger pixels of the 600D have less shot noise. My idea was to compare the noise at the same resolution. When it comes to presenting the image on screen which is 4k at best or more often only fullHD both images are scaled down anyway. There was neither deconvolution nor denoise done. Both images needed some crop to get rid of the bad borders from dither. As the filter characteristic is very likely different I did a photometric color calibration. For stretching the values suggested by PI's automatic screen stretch were transferred to HistogramTransformation. This results in a strong stretch with a far too bright background. To get rid of an possible intensity differences a linear fit was applied to the 800D's image using the elder sister as a reference. Here is an animated gif, toggling between the two images each second. There is a surprise:
Pleiades, Canon EOS 600D(T3i) vs. Canon EOS 800D(T7i)
The halos are smaller in the 800D's image. As I swapped cameras every 16 minutes over the course of the night this cannot be caused by clouds, fog or focus. It must be the sensor. I was wondering if this is the result of almost seven years of dust collecting but I clean the sensor now and then. My interpretation is that the new 24MPix sensor has got a less aggressive anti alias filter. It also seems like the filter package has got a better anti reflective coating. Have a look at the 100% crop of Alcyone in the middle of the Pleiades, again toggling between both cameras:
Alcyone at full resolution, 600D vs. 800D
Two major differences to observe: first, the noise level is lower. It is not mind blowing but clearly lower. Second, I can see a relatively sharp border of the halo in the 600D's image while in the 800D's the halo is smooth and smaller.

Conclusion

For my level of light pollution and for this bright object the 800D performs better. If the change in the halos is systematic to the sensor or filter package and not a result of dust on the old sensor the 800D will produce better images in any case, even if the background level or the temperature is different. One minor point to complain about is the change of grid lines in life view. They used to be gray in the 600D, now they are black which is perfect for daylight but not at night. Turning the display to a high level and the ISO to maximum generates enough noise to make the lines visible again. If you do not align your mount to the life view but plate solve or use an eye piece, it's not a problem at all. The 800D is a few grams lighter but not really much. If you are on a budget I can still recommend the 600D for the start. I'm not the kind who starts complaining how incredible bad something is because a new slightly better model is out. Both are good cameras for astro photography. As you proceed to dimmer object the 800D is the better option.

Thanks for reading!

Remark: The exact same data have been processed again without dark frames. Read more about it here: Are dark frame overrated?

Back to Media