To Harry and ALL,

I have been using a Canon EOS Rebel XTi (same as EOS KISS X - funny name in Japan). I will call it “X”. Focusing through the viewfinder has been difficult. I bought a Canon EOS Rebel T1i, same as EOS KISS X3. I will call it “X3”. Its capability of high ISOs such as 12800, 6400, so forth, live-view and video, was quite attractive. The old Rebel XTi was given to my son. I have not yet used the video much and hope to use it in occasions such as the crash onto the Moon in the LCROSS project. The live-view in the new camera improved the quality of my photos “astronomically”.

Well, now there are some street rumors about the high noise with T1i. However, the noise evaluation data is sketchy. Therefore, I “borrowed” the old Rebel back from my son and performed a test, comparing the new Rebel against the old, literally side by side.

Based on Experimental design, the factors and the levels selected are:

Ambient temperature for camera: 7, 17 and 29 º C
ISO: 800 and 1600
Exposure seconds: 30,180,300 and 600 seconds

The dark images are plotted and the mean/standard-deviation of the RGB values (0-255 scale) is obtained (Photoshop Elements 7 is used). The mean noise RGB value will be called “MNRGB” from here on.

Summarizing the results:

1. There is a critical combination only below which X3 is better than X and above which X3 is worse than X. The combination is around 17 º C, 180 seconds, 1600 ISO. If, for example, one combination has 25 º C instead of 17 º C, X3 is worse. If, for another example, it has 300 seconds instead of 180 seconds, X3 noise is worse. In other words, X3 has a lower noise in the low noise region of low temps, low ISO and low seconds, when compared to X. X3 is worse at the higher noise region.

2. MNRGB (mean noise RGB value) of both X3 and X stay below 2 (0-255 scale) in the low noise region. When one of or all of the three factors; temps, ISO and seconds, increase, the MNRGB increases more rapidly in X3 than in X, reaching 60~80% higher noise, a rather poor temperature characteristics with X3.

3. In this paragraph, noise and signal levels are discussed.

I ran the dark noise exposures for the same combination twice with 1 hour separation. Subtracting the #1 bitmap from #2 bitmap, I obtained roughly 50% reduction of the noise.

I am a beginner and doubted text book writing, explaining the effect of dark noise reduction. Because the noise is “thermal” and, as the result, is random, where the noise reduction can be done only by averaging many images. But, It seems I was wrong and had to search some CCD related papers. I found ones written by Hamamatsu Photo electronics and Sony. My goodness! The involving physics was related to my area in the ancient time. So, I forgot most of them.

Mainly, the surface defects of semiconductor produce the surface energy levels from which the electron excites due to photon, creating the noise. Note that when the electron is excited from the bulk area, the excitation is normal. Therefore, the high noise intensity at various spots is originated in the manufacturing process and hence we can reduce the noise simply by subtracting a few, not many, averaged dark noises from a few averaged images. Since the image subtraction reduces the noise by 50%, I expect we can reduce the noise down to 1/3 to 1/4 if I do it statistically and carefully. I may be too optimistic, though.

This discussion leads to a conclusion that:

I will keep X3 and using it. Probably I will be able to take a relatively faint deep sky photos with up to the max combination; 10 minutes exposure at 1600 ISO at 17º C, typical California night temps. Although MNRGB is large, 12, with this combination, I will be able to reduce the MNRGB down to 3~4, using the dark subtraction.

4. Based on the noise, evaluated in #3, I estimated the RGB value of the rather faint clouds away from the Galaxy center, in M31, Andromeda, on my unprocessed image (300 seconds exposure, 800 ISO, Sigma 500mm Zoom of 80mm diameter, F/6.3, Orion skyglow filter). It was around 10, in 0-255 scale, above the background. Of course, the galaxy center was 154 above the background. This could give the RGB value of 40, as a signal (=10/5 x 1600/800 x 10), if I use the max combination. Because the noise level is 3~4, I can live with this scenario. But, correct me if my calculation of the nebula intensity is wrong.

5. The data shows a rather severe temperature effect in X3. Certainly I do not recommend X3 to my friends in New Orleans and Ghana, West Africa. Probably, X3 is okay with my Canadian friend if he forgets about three hot nights in one summer. For my fellow Californian hobbyists, the choice would be a little complex, probably depending on his need for the video capability.

I do not have any data about “KISS X2” which seems a recommendation of some of the scope shops in Japan. It is said to be between X3 and X.

6. Additionally, the data shows that high ISO & short exposure is less noisy than low ISO & long exposure, suggesting a possibility of “point-and-shoot” deep-sky astrograph. But, a further test using the high ISO will be required to confirm the fact. Please note that grainy surface with high ISO is not fully related to this thermal noise, I believe.

7. One more additional finding is that the camera is heated up internally during the long exposure and, as the result, the noise generation increases by almost two fold for X3. A good preparation and image taking right at the beginning, before heating-up, is favored. Variable angle LCD display from Nikon could be effective to a certain extent because the LCD panel and the power delivery system generates a lot of heat.

The RGB values are shown in:

http://www.urata123.com/X-urata/astro/EOSNoise/DarkNoiseComparison.pdf
The images of the dark files are attached in:

http://www.urata123.com/X-urata/astro/EOSNoise/DarkNoiseComparisonGraphs.pdf

A comparison is extended to Nikon D90 and D5000, attached in (this link is to be activated when a permission of the D90 and D5000 data generator is received):

http://www.urata123.com/X-urata/astro/EOSNoise/EOSvsDXX.pdf

The conclusion in this extension is probably that D90 and D5000 are par with X3 and X, respectively. The data is not as reliable as it needs to be and thus will be presented as just one reference.

Nobi