Canon 1000D comparison to 450D

[RobertM]

I now have a Canon 1000D bought off EBay which I aim to modify and use purely for astrophotography.  Before I modify and bugger it up I'm going to perform some basic tests - read noise/bias/darks etc. as I think it would be useful for those considering using one if these for the same purpose.

All noise reduction turned off etc, etc...

Notes frame sizes and download times:

EOS1000D: 3906 x 2602 pixels, CR2 file size 8,578kb, download time approx 2s
EOS450D  : 4290 x 2856 pixel, CR2 file size 11,317kb, download time approx 3s

Both cameras set to ISO 800

Bias frames:

20 x 1/4000s taken at 1 intervals

Taking the 10th frame from each camera, just straight conversion in MaximDL with no scaling.

EOS450D
              Red     Green   Blue
Maximum:  2076    901    1370
Minimum :       0    518        0
Average :  1307     696   1061
Std Dev :   26.0     8.2    21.0
 
EOS1000D
              Red     Green   Blue
Maximum:   418     222     361
Minimum :      0     125        0
Average :   326     174     264
Std Dev :    7.0     2.3      5.6

These are figures for the whole frame and whilst they don't tell us that much I can already see that the standard deviation of the noise on the 1000D is significantly lower.  This may change when I do the dark frames over the weekend.

Next installment tomorrow will be read noise.

Robert

[Carole]

Very interesting Robert.

[MarkS]

Robert,

Those figures are in ADUs.  To make a meaningful comparison they both need to be measured in electrons.  But you'll end up doing this anyway to calculate the read noise.

http://forum.orpington-astronomy.org.uk/index.php?topic=3473

Mark

[RobertM]

Updated the original post with ISO.

Thanks for the feedback.

[Mike]

Interesting. Thanks Robert. The 1000D is one of the cameras I have been considering.

[Carole]

No hope for me then, I have a 450D

[RobertM]

It's purely a comparison Carole, I'm not sure which camera will work out better.

[mickw]

Strange the 1000 is reviewed as a detuned 450 or a tuned 400, sensor (mp) being the same as 400 but less than 450.
The sensors are the same physical size so the the 1000 has larger pixels than the 450 ?
That would suggest that bigger is better (in this case)

[RobertM]

EOS 1000D pixels are 5.68 micron = 32.3 square microns
EOS 450D pixels are 5.15 micron = 26.5 square microns

You would think a small but significant difference however a lot is down to the sensor design.  For example the 550D is supposed to be much better at low light than the 450D despite having tiny 4.17 micron pixels.

[mickw]

I read somewhere something along the lines of the lenses (?) being smaller allowing the pixels to be stacked closer on certain sensors.  If that's the case then there would be less dead space and more area assigned to actually catching photons - I think

*Edit*
Where I read it -
http://www.dpreview.com/reviews/canoneos550d/

[Carole]

I'm afraid some of this technical stuff is beyond me.  When you've finished Robert would you mind posting your results in non technical terms as well please.

Thanks

Carole

[RobertM]

Well, I've finally got around to doing some measurements and am not getting the results I expected.

Firstly the saturation point of each of the colour channels is at a different ADU level:

           R           G            B
450D    19014    10120     15429
1000D    4452     2370       3614

Remember the 450D has 14bit A-D converter whereas the 1000D has 12bit and because of that you would need to multiply the 1000D figures by 4 to give a 16bit equivalence.

This was measured using a very simple procedure.

1) Take a flat field image long enough to saturate each colour across the ccd (10s in this case)
2) Import to MaximDL and subtract the master bias (10 frames)
3) Convert to colour using the conversion routine for the camera model.
4) Measure the average pixel value in ADU's.

Because of this I thought it would be worthwhile calculating the gain and full well depth of each colour and these are the results.  Intrigueingly the channels seem to be quite different when I would have expected them to be the same.  There seem to be two possible explanations for this:

1) There are different gains for each channel.
2) I've made a mistake !

Working out the figures assuming I've done everything correctly:

Gain
===

450D
-----
Channel   S   Sigma     Noise   Variance   Gain (e/ADU)   Full well ADU's   Well depth (e)
R        2603   76.45        54.06   2922.88   0.89                  19014             16933
G        2358   42.48        30.04   902.42   2.61                  10120            26443
B        3259   68.31        48.31   2333.63   1.40                  15429            21547

1000D
------
Channel   S   Sigma    Noise    Variance   Gain (e/ADU)   Full well ADU's   Well depth (e)
R          874   19.05    13.47    181.51   4.82              4452             21438
G          768   10.56    7.47      55.73    13.78             2370             32660
B          1058   17.16    12.13    147.24   7.19               3614             25968

The full well depth is just calculated by multiplying the gain by the saturation values above.  The Gain is the number of electrons that represent one ADU (Analogue to Digital Unit).

All measurements made at ISO 200 in each camera.

On the face of it the 1000D has a higher well depth than the 450D but each ADU represents more electrons so there is less resolution in the signal.

Thoughts anyone ?

Next Step - Read noise.

Robert

[MarkS]

Robert,

The gain for each channel should be the same and the well depths (in electrons) should be similar (though not identical because of the bias).

I think that colour balance scaling is being applied at some stage in your processing and this is upseting your calculations.  An RGB colour balance multipliers of (1.9, 1.0, 1.5) are broadly correct for a "sunny day" colour balance.  This is what you appear to be seeing with the RGB ratios in your saturated frames.

Mark

[RobertM]

Thanks Mark

I was mulling that over when I wrote the message last night but then thought that the RAW's would not be affected by colour balance.  I wonder whether MaximDL is applying some correction even though I made sure the scaling factors were 100% (ie. no change).  This behaviour could explain a few things.

Will do some more digging...

Robert

[RobertM]

I'm away at the end of the week so spent this evening removing the IR cut filter from the 1000D.  It's now much more sensitive to IR.  I still have all the figures from before the mod so will continue the calculations when I get the chance.

Robert