Oddness with AV - General Discussion and Assistance - CHDK Forum
supplierdeeply

Oddness with AV

  • 11 Replies
  • 3895 Views
Oddness with AV
« on: 11 / November / 2009, 10:55:47 »
Advertisements
I hadn't previously taken note that the AV bottom limit changes with zoom factor, I did all my testing at full wide.
Anyways did some testing in it and found some oddness.  At full wide, the aperture was limited in CHDK to F11, any settings beyond were ineffective (acted like F11).  I had dug through the firmware and found it was getting software limited (http://chdk.setepontos.com/index.php/topic,4393.0.html), and was even able to override this limit.

Anyways, retesting the unmodified CHDK at full zoom, I found that F16 was working (still registered as F11 though) which I thought as odd as there was a software ceiling set to F11.  So I did some further digging.
What I found was that the propcase apex96 values for AV were the same for the same AV at both wide and zoom (e.g for F8.0, av96=576) however, at the entry point in the FW where it gets converted to another value (as outlined in the thread above), the values were quite different.  For full wide, they value going in was the same av96 (F8.0 =576) but for full zoom, it was different (F8.0 =382).  For every av value it was off by the same 194 units, and the converted value being pulled in was the one for this value.

First, this explains why F16 works for zoom but not wide, the conversion table that it pulls the value from was limited to F11 (av96=672, value = 1903).  At full zoom, the F16 av96=574 (value=1533) which is still within the conversion table limits.  In fact (I haven't got to this test yet) this should mean that at full zoom, with the base CHDK, aperture should be able to be set as high as F22.8 (equivalent of wide angle F11, av96=672) and that with my previously modified override, it may be able to set it as high as F32.2 (equivalent of wide angle F16, which seemed to be the upper limit I could get).

I suppose that this has something to do with how aperture values work (ie, the same aperture AV value for wide vs zoom, does not have the same actual physical aperture opening = at a larger zoom, the physical aperture opening is wider for the same AV value).  I suppose this may actually make sense for someone who understands camera optics a bit better (time to do some googling and learning on this maybe).  Perhaps it is because the lens unit is futher from the aperture location?

Now I need to find how the zoom value and the actual aperture value relate to each other (at full zoom, av96*=av96-194).

The actual physical limits (or limits I haven't tracked yet) for the final av96* still appear to be ~292-768 (F2.8-F16.0 at wide, F5.6-F32.2 at zoom), but I will still need to do some verification on this.

Will post more information as I find it.  If anyone has further insights on this, please let me know.

Re: Oddness with AV
« Reply #1 on: 11 / November / 2009, 12:09:07 »
For every av value it was off by the same 194 units

Which is fractionally more than two stops.
The SX10 fully open aperture varies from 2.8 to 5.7 as you zoom, fractionally more than two stops !
The aperture value is determined by focal-length and the apparent diameter of the aperture as seen in its image when viewed from the front of the lens.
As the lenses in front of the aperture change position as you zoom, so does the image-size of the aperture.
There f# varies.

Quote
it may be able to set it as high as F32.2

With all due respect, you cannot be serious ?
CHDK hackers try to push the limits Canon have imposed without understanding why those limits exist.
(the same goes for absurd claimed shutter-speed overrides).

Quite simply, f32 is equivalent (in terms of depth-of-field) to f180 on a 35mm film camera !!!!

Show me a regular  35mm film camera with that aperture.

Due to diffraction, image resolution would be terrible.
« Last Edit: 11 / November / 2009, 12:31:48 by Microfunguy »

Re: Oddness with AV
« Reply #2 on: 12 / November / 2009, 09:44:16 »
I think you may be missing what I am saying here.
I know that there are physical hardware limits on the aperture opening size (it can't physically open or close any further) and that there are software set limits.

For the SX10, at wide angle there are 3 limit cases:

1. Non-CHDK gui: F2.8-F8.0
2. CHDK (current): F2.8-F11.0
I found this second limit was still being software limited by the FW and made a workaround and test compile:
3. Modified CHDK: F2.8-F16.0

From my testing it would appear that the extreme hardware limits are equivalent to about F2.8-F16.0 as pushing further does not yield different results (doing a direct comparison of image exposure level).
This corresponds to av96 values of 288-768.
The FW converts these values into a second value, which I assume is used for setting the aperture mechanism (I don't know what the actual significance of the value is, but they are *mostly* linear).

Now as we zoom out, the bottom aperture limits increases.  As you have confirmed, this is basically due to how AV is calculated, it is a factor of both the focal length and the physical opening size of the aperture.  Because of this, and because there is a bottom aperture *size* limit (av96 288 @ wide), this would make sense.  At the largest aperture opening size at wide, the AV is F2.8, at full zoom for the same opening size the AV is F5.6.
What happens in the firmware is that at full zoom the av96 value for the current AV gets stepped down by 194 (eg F8.0 ---> av96 = 576 - 194 = 382) and this value is used to set the aperture size (ie F8.0 (576) at full zoom has the same aperture opening as F3.97 (384) at full wide).  This offset value is constant for all AVs for a given zoom factor (zoom=0 (wide) factor = 0, zoom=128 (full) factor = 194, zoom=97 factor = 125, etc.).

I wanted to know the relationship between zoom factor and the AV offset factor, however this proved unnecessary, as the value was available in the FW code where I needed it. (For curiosity, I would still like to know the exact relationship.  I know it isn't linear based on the values I have got: 0=0, 41=43, 97=125, 128=194.  Eventually I may try to collect all 129 values and check the exact relationship.)

Now this offset has further implications when we look at zoom and it's aperture/AV relationship. At full zoom, the AV gets limited to F5.6-F8.0.  This is equivalent to the aperture size for F2.8-F3.97 at wide or 288-382. However, we know the aperture is physically limited to 288-768, so we should be able to push the aperture up to 768 at full zoom.  This would be equal to F32.2 (768+194=962 -> F32.2), but only at full zoom.  At other zoom levels, the top and bottom limits would calculate out differently based on the offset value:
z=  0   (0): F2.83-F16.00
z= 41  (43): F3.30-F18.69
z= 97 (125): F4.44-F25.12
z=128 (194): F5.70-F32.23

I am not saying F32 is available at wide angle!  That would be impossible! (For the given hardware...)


Hopefully my rationalisation is understandable here.

But outside all these theoretical numbers, the question would be is it true?  And the answer is yes.  I made some test modfications to the firmware.  The values I needed were all available:
[SX10 1.03a] Upon return from the conversion function (FFA568A8) at FF93B4E8, register R0 holds the converted value (aperture setting), R5 holds the offset av96 value, R6 holds the offset value.
Using these values, I convert the av96 into the corresponding aperture setting and save it to R0 where necessary.

Testing the resulting exposure levels using this build was as expected:
At full wide:  Exposures varied from F2.8 up to F16.0
At full zoom:  Exposures varied from F5.6 up to F32.2

The last issue here, is whether or not it is usable.  I had previously tested up to F16.0 at full wide and found a little diffraction at F11 and even more so at F16 (however these were at close-up shots ~12in).

I will still be trying some tests at full wide and zoom to see what the diffraction level is.


For every av value it was off by the same 194 units

Which is fractionally more than two stops.
The SX10 fully open aperture varies from 2.8 to 5.7 as you zoom, fractionally more than two stops !

I know this which was my point.  Why is F2.8 @wide = F5.6 @zoom?  I'm sure there is some calculatable function for this based on aperture size and focal distance.
As I noted above, it's not necessary to know the relationship as the offset value needed is readily available in the FW.


it may be able to set it as high as F32.2

With all due respect, you cannot be serious ?
CHDK hackers try to push the limits Canon have imposed without understanding why those limits exist.
(the same goes for absurd claimed shutter-speed overrides).

Yes I am serious, as outlined and tested above.  Be clear on what I am referring to here though: the F32 AV is at full zoom only, not at full wide.


I am aware of the difference in being able to tell the camera to do something and it actually doing it.  There are hardware (and software) limits on everything.  However what I have stated here does not extend beyond what the know hardware limitations are, it just removes some of the software limits.
I have noted people stating ridiculous shutter speeds just because they can set it, without actually verifying it.  Even with the SX10, on one of the Flickr groups, people were initially noting some pics taken at very high shutter speeds, whereas, I had checked and found that this was likely not actually happening as there were distinct software limits built into the FW that CHDK wasn't bypassing. (http://chdk.setepontos.com/index.php/topic,4392.0.html).  Fixing this, I was able to test and verify shutter speeds up to ~1/16000-1/32000 which is surely not adsurd if it can be proven.


Due to diffraction, image resolution would be terrible.

This is true and the bottom line on the usefeluness of all this.  However, in the end, the usefulness is in the eye of the photographer.  Someone may see or have a use for the expended AV settings, regardless of the diffraction level.


Re: Oddness with AV
« Reply #3 on: 12 / November / 2009, 10:13:19 »


Re: Oddness with AV
« Reply #4 on: 12 / November / 2009, 11:43:10 »

Re: Oddness with AV
« Reply #5 on: 12 / November / 2009, 12:44:23 »
You have merely shown that above Canon's shutter-speed limit there is a progressively-smaller change of exposure.
These are not proper shutter-speeds in the photographic sense.
We already known from tests published on the WIKI that exposure changes by some small amount.

Within the cameras normal range, each doubling of shutter-speed halfs the exposure.

Even at an alleged 1/4000 sec that starts to fail.

If you wish, you could relabel your 1/64,000 to 1/14,000 sec, that is more likely based on the figures.

Your 1/16,000 is a bit better at a real 1/10,000 sec .. if your data is correct.

Even then they are of no practical use, even 1/2000 sec at 400 ASA  on a bright sunny day is too fast !


Re: Oddness with AV
« Reply #6 on: 12 / November / 2009, 13:23:48 »
This discussion is useful because you have given me an idea.
In SDM I will remove the highest shutter-speeds that can be used as overrides.

I will replace by just a few (maybe to 1/15,000 sec) that are true shutter speeds in terms of halfing of exposure.
That will mean doing tests and replacing the calculated Tv96 values in the table with the values obtained from experiment.

Re: Oddness with AV
« Reply #7 on: 12 / November / 2009, 14:16:10 »
This discussion is useful because you have given me an idea.
In SDM I will remove the highest shutter-speeds that can be used as overrides.
I will replace by just a few (maybe to 1/15,000 sec) that are true shutter speeds in terms of halfing of exposure.
That will mean doing tests and replacing the calculated Tv96 values in the table with the values obtained from experiment.
It is quite difficult to get an accurate shutter speed, unless you can test on a moving object of known speed where distance can be noted (eg dremel tests, however, you would need to accurately know the rotation speed).
That is the reason why I've noted that exposure tests are not meant to be accurate.  One issue that arises is how noise affects the exposure.  Usually, the pixel values are quite high relative to the noise level, however, at the exposure levels in the ones I did for the tests, noise can have a large effect.
For example, the average pixel values for the 1/8000 is 0.1173.  A very rough estimate of the average noise level can be around 0.01 or higher (again this is very rough estimate, meant more as an illustration), which would lower the real average pixel value by around 10%, as the exposure level gets lower, this effect can become much more pronounced.


Re: Oddness with AV
« Reply #8 on: 12 / November / 2009, 15:14:49 »
I will use a light source so bright that at 1/1000 sec the hitogram does not quite reach 255.

I will only be reducing the exposure by 16x (four stops).

I have a CSI light source for a microscope that is so bright I only dared turn it on once !

Re: Oddness with AV
« Reply #9 on: 13 / November / 2009, 16:59:12 »
I will use a light source so bright that at 1/1000 sec the hitogram does not quite reach 255.

I will only be reducing the exposure by 16x (four stops).

I have a CSI light source for a microscope that is so bright I only dared turn it on once !

I think I may be looking at my exposure comparisons a bit wrong or what am I missing here?  I wanted to check how it works so I did a series of photos similar to how I did above, but for normal shutter speeds and apertures.  I tried to make sure the exposure range avoided blowing out white and bottoming out black as much as possible to try to get as close a comparison as possible and checked at some 1 stop intervals (1/8s & 1/15s, 1s & 1/2s, etc) and checked the average pixel values, expecting the relative ratios to be close to 0.5 (half brightness, full stops), however they came out around only 0.73-0.75 on average.  This was all done loading the images at a linear (1.0) response curve which I assumed should be the proper calibration.  I could also note, the the overall histogram shift was around this magnitude vs the 0.5 I expected.  I then noted that applying the standard 2.2 response curve, the exposure ratio seemed correct now (note 0.73^2.2=0.5).  So should I be comparing my exposures at all levels using a 2.2 response?

(We should also maybe move this discussion to a new thread so that others may notice and get involved, and this thread can remain on focus with the original AV override topic)