Ad
Ad
Ad
Pages: « 1 ... 5 6 [7] 8 »   Bottom of Page
Print
Author Topic: Mark Dubovoy's essay  (Read 32727 times)
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3418


« Reply #120 on: November 10, 2010, 11:08:49 AM »
ReplyReply

A few comments.

1. Stops are a logarithmic/power scale, so 70% is half a stop (as 50% is one stop). A one stop decrease in illumination is a factor of 1/2 in light, so a half stop increase is the square root of that, or sqrt(0.5) or about 0.7.

You are right. Must have hit the wrong calculator button (Log(0.7) / Log(2) = -0.51 stops).

Quote
I could dig up a Dalsa reference on microlenses reducing acceptance angle if you want.

That's not necessary, I can find it if needed. There obviously is a limit to the acceptance angle of a microlens, but micro-lens technology has also advanced (e.g. offset micro-lenses (not good for shifted lenses) and different shapes and refractive indexes). Also, Mark's essay showed DxO charts with maximum size 24x36mm sensors, so the maximum angle of incidence near a corner is more limited than on a MF sensor.

In the center of the sensor array (as in the DxO measurements) most of the exit pupil rays strike the sensels almost perpendicular, and those from the edge of the exit pupil will be refracted to something closer to perpendicular. So whatever DxO measured, it was almost certainly not impacted by shading effects.

Cheers,
Bart
Logged
BJL
Sr. Member
****
Offline Offline

Posts: 5120


« Reply #121 on: November 10, 2010, 03:41:37 PM »
ReplyReply

There obviously is a limit to the acceptance angle of a microlens, but micro-lens technology has also advanced (e.g. offset micro-lenses (not good for shifted lenses) and different shapes and refractive indexes).
Of course the technology is advancing, but the evidence of quite recent Kodak FF CCDs with microlenses is that they still have far more severe off-perpendicular sensitivity fall-off than sensors without (30% loss or 1/2 stop at 27º with, vs. 10% at 27º and 30% at 40º without.) And off-setting is irrelevant to the problem I am talking about which applies even at the center; off-setting is only relevant to the additional problems with wide-angle (low exit pupil) lenses, an issue that seems of little importance with DSLRs when used with SLR lenses, due to modern mainstream SLR lenses having fairly high exit pupils relative to sensor size.

Perhaps you are thinking that somehow Kodak is behind what Canon, Sony etc. are achieving with their CMOS and ILT CCDs, but the opposite is probably true: CMOS and ILT CCD sensors require stronger microlenses than FF CCD (the wells are a smaller portion of the total photosite) making the problem worse, and it is yet worse for CMOS sensors (front-illuminated ones anyway as all current DSLR CMOS sensors are) as the stack of transistors on top of the photosite forces the microlenses to be further from the well. Some quotes from Dalsa page 7 of http://www.dalsa.com/public/dc/documents/Image_Sensor_Architecture_Whitepaper_Digital_Cinema_00218-00_03-70.pdf

- about ILT CCD in particular, but true in general:
At low f-numbers, microlensed pixels can suffer from vignetting, pixel crosstalk, light scattering, diffraction and reduced  MTF—all of which can hurt their resolving power.

- about T3 CMOS in particular, and even more so of T4/T5 CMOS:

The tradeoffs involved with microlenses are more pronounced with CMOS imagers since the microlenses are farther from the photosensitive surface of the pixel due to the “optical stack” of transistors.


In the center of the sensor array (as in the DxO measurements) most of the exit pupil rays strike the sensels almost perpendicular, and those from the edge of the exit pupil will be refracted to something closer to perpendicular.
Dalsa disagrees on that last point, indicating instead that the already oblique rays are bent even more off-perpendicular. See the top-center figure on the same page 7 as cited above. Or just note that microlenses are converging lenses.

This illustrates the potential problem: light from the edge of a low f-stop light cone, at any part of the sensor, can be sent off to the side of the well, and be either not detected or get smeared into an adjacent well. Kodak's masking is to avoid the latter.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8849


« Reply #122 on: November 10, 2010, 07:38:11 PM »
ReplyReply

Highly entertaining! I'm glad we are beginning to see the results of some test images at shallow DoF, but I can't help being reminded of the following amusing parable.  Grin


Quote
In the year of our Lord 1432, there arose a grievous quarrel among the brethren over the number of teeth in the mouth of a horse. For thirteen days the disputation raged without ceasing.

All the ancient books and chronicles were fetched out, and wonderful and ponderous erudition such as was never before heard of in this region was made manifest.

At the beginning of the fourteenth day, a youthful friar of goodly bearing asked his learned superiors for permission to add a word, and straightway, to the wonderment of the disputants, whose deep wisdom he sore vexed, he beseeched them to unbend in a manner coarse and unheard-of and to look in the open mouth of a horse and find answer to their questionings.

At this, their dignity being grievously hurt, they waxed exceeding wroth; and, joining in a mighty uproar, they flew upon him and smote him, hip and thigh, and cast him out forthwith. For, said they, surely Satan hath tempted this bold neophyte to declare unholy and unheard-of ways of finding truth, contrary to all the teachings of the fathers. After many days more of grievous strife, the dove of peace sat on the assembly, and they as one man declaring the problem to be an everlasting mystery because of a grievous dearth of historical and theological evidence thereof, so ordered the same writ down.
« Last Edit: November 10, 2010, 07:39:44 PM by Ray » Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3418


« Reply #123 on: November 11, 2010, 06:06:23 AM »
ReplyReply

Of course the technology is advancing, but the evidence of quite recent Kodak FF CCDs with microlenses is that they still have far more severe off-perpendicular sensitivity fall-off than sensors without (30% loss or 1/2 stop at 27º with, vs. 10% at 27º and 30% at 40º without.)

I don't think that applies in general, while it does in this specific case where one sensor design only uses a small part (< 1/2 of the sensel pitch) of the surface to collect light, and the other half as signal buffer). Besides, compare that performance with the supposed Cos(a)^4 light fall-off (e.g. on film).

Quote
Perhaps you are thinking that somehow Kodak is behind what Canon, Sony etc. are achieving with their CMOS and ILT CCDs, but the opposite is probably true: CMOS and ILT CCD sensors require stronger microlenses than FF CCD (the wells are a smaller portion of the total photosite) making the problem worse, and it is yet worse for CMOS sensors (front-illuminated ones anyway as all current DSLR CMOS sensors are) as the stack of transistors on top of the photosite forces the microlenses to be further from the well.

No, I'm not thinking Kodak is behind the curve, there are more Kodak sensors being used in all sorts of equipment than most people know. It's a very mature division within that company.

Quote
Some quotes from Dalsa page 7 of http://www.dalsa.com/public/dc/documents/Image_Sensor_Architecture_Whitepaper_Digital_Cinema_00218-00_03-70.pdf

- about ILT CCD in particular, but true in general:
At low f-numbers, microlensed pixels can suffer from vignetting, pixel crosstalk, light scattering, diffraction and reduced  MTF—all of which can hurt their resolving power.

- about T3 CMOS in particular, and even more so of T4/T5 CMOS:

The tradeoffs involved with microlenses are more pronounced with CMOS imagers since the microlenses are farther from the photosensitive surface of the pixel due to the “optical stack” of transistors.

Dalsa disagrees on that last point, indicating instead that the already oblique rays are bent even more off-perpendicular. See the top-center figure on the same page 7 as cited above. Or just note that microlenses are converging lenses.

Maybe you misread my remark, I was specifically talking about the center of the sensor array. The only obliquness there is from the marginal rays from the exit pupil, the rest around the chief ray goes in almost perpendicular. Nobody contests that in the corners of the sensor array the task for microlenses is more daunting. Looking at the DxO graph in Mark's essay, CMOS doesn't seem to do that bad in the center of the sensor array, compared to CCD (although there are more variables in play that could cause that).

Quote
This illustrates the potential problem: light from the edge of a low f-stop light cone, at any part of the sensor, can be sent off to the side of the well, and be either not detected or get smeared into an adjacent well. Kodak's masking is to avoid the latter.

Yes, manufacturers are very well aware of optical cross-talk, and take precautionary measures, e.g. by varying the thickness/shape/and refractive index (which can also influence internal reflection), and adding exposure masks, to reach a better overall optimum, as demonstrated here:
http://www.lumerical.com/fdtd_microlens/cmos_image_sensor_pixel_microlens.php

And here is another interesting paper (published in 2000, it may be a bit dated, but it shows the considerations nicely):
http://www-isl.stanford.edu/groups/elgamal/abbas_publications/C074.pdf

Although it deals with sensors without microlenses, they do add at the end of the document:
"Microlenses: QE can be increased using microlenses.The design and manufacturing of these
microlenses must, however, take into consideration the geometry of the tunnel. The improvement in
on-axis QE can be considerable. For off-axis pixels, it is not clear how much improvement, ifany, is
achieved without the use of a telecentric imaging lens or an unconventional angle-dependent microlens
design, which may not be practical. Another important consideration is the material needed. To
focus the light onto the photodiode surface, the focal length of the microlens may be too long to be
manufactured using silicon dioxide. In this case, other materials such as polymethyl methacrylate
(PMMA) may be used, which increases the cost of the microlens fabrication.".

Cheers,
Bart
« Last Edit: November 11, 2010, 09:00:50 AM by BartvanderWolf » Logged
pegelli
Sr. Member
****
Offline Offline

Posts: 581



WWW
« Reply #124 on: November 11, 2010, 11:12:37 AM »
ReplyReply

I don't have a 5dII and a 85/1.2 L lens, but I do have a Sony A850 and an older Minolta AF 85/1.4 G lens (very nice combo for portraits)

So I decided to test and see if the circle of confusion at f 1.4 became smaller as expected by comparing it to the circle of confusion at f2.8.

Here's the three images:
Sharply rendered highlight:


out-of focus @ f2.8:


out of focus @ f1.4:


Since the Minolta 85/1.4 has some internal focussing I made the out-of-focus shots at the same distance setting (~1 meter) but with a 12 mm extension ring added. So I know for sure all 3 shots are taken at exactly the same focal lenth.

My conclusion is that the diameter ratio of the circles of confusion between the f1.4 shot and the f2.8 shot is exactly 2, so the out-of-focus and dof rendering at f1.4 is exactly as expected. So the dof reduction effect and shielding of marginal rays at large f-stops as hinted at by Mark Dubovoy does not show up with this camera/lens combination.

Unfortunately no f1.2 A-mount lenses were ever produced, so we'll never know how that would work. There are however also a 35/1.4 and 50/1.4 for the A-mount but since I have neither I cannot test those at this moment.

I'll do some tests later to see if I can find any "ISO" gain as Mark is describing with this combo, but since that interests me much less it will have to wait until I make some time for that.
« Last Edit: November 11, 2010, 12:07:21 PM by pegelli » Logged

pieter, aka pegelli
BJL
Sr. Member
****
Offline Offline

Posts: 5120


« Reply #125 on: November 11, 2010, 11:34:37 AM »
ReplyReply

I don't think that applies in general, while it does in this specific case where one sensor design only uses a small part (< 1/2 of the sensel pitch) of the surface to collect light, and the other half as signal buffer).
The Kodak sensors compared are both full frame CCD's, so as far as I can tell they have no signal buffer, just a little space at the edges for lateral overflow drains ... and one of the main claims in that Kodak paper is that the design reduces the space needed for the LODs, giving an impressively high 69% of the photosite being electron well (some is then masked off from direct illumination, but can still hold electrons). This loss to LODs is less than with the CMOS or interline CCD designs in most SLRs.

To repeat, the need for a smaller "window" (more masking) over the well in the microlens design is forced by the difficulties like pixel cross-talk that are made worse by microlenses; it is not some dumb stupid thing that Kodak did with the KAF-31000, degrading its performance in one respect, while knowing how to to better in another (the KAF-39000 without microlenses).

Since I am the only one providing actual data about actual sensors and quotes from sensor making companies, can you provide authoritative evidence to the contrary, like spec's a sensor with microlenses than has off-perpendicular sensitivity "at the center of the sensor array" as good as in that non-microlens sensor? You might find one such, a new one from Dalsa using its new "low profile" microlens design, but that approach only works with CCDs, not CMOS, as Dalsa itself explains, so I doubt that any recent mainstream SLR sensor (all CMOS with microlenses) has such performance.

Quote
Maybe you misread my remark, I was specifically talking about the center of the sensor array.
The only obliquness there is from the marginal rays from the exit pupil, the rest around the chief ray goes in almost perpendicular.
No, I understand it perfectly: as I have said repeatedly, we are both talking about an effect that applies to marginal rays at low f-stops everywhere on the sensor, including at the center of the sensor where the chief ray is perpendicular but marginal rays are substantially off-perpendicular. And the illustrations that I referred to in the Dalsa document, the middle column, are exactly for this situation.

Quote
Yes, manufacturers are very well aware of optical cross-talk, and take precautionary measures, e.g. by varying the thickness/shape/and refractive index (which can also influence internal reflection), and adding exposure masks, to reach a better overall optimum ...
And from here on we agree, and so do your sources, neither of which in the slightest contradict my point that light loss also occurs at the center of the sensor array with low f-stops. The problems increases away from the center, and only produces "vignetting" (variation in detected luminosity detected de to spatial variation in QE) off-center, and so the off-center effect is discussed more but produces somewhat reduced QE everywhere, even at center, with sufficiently low f-stops. Also no MF lens has an f-sop low enough for this to be an issue, whereas many have low exit pupils, and Kodak explicitly says that these designs are optimized for MF, so the low f-stop effect is not worth Kodak's discussing in that document.

Until you come up with superior authority or arguments, I have to go with Kodak's statement that
the KODAK KAF-39000 Image Sensor (39 Mp) is designed without microlenses to maximize incident light-angle response ... the critical crosstalk angle is increased
and of the KAF-31000 that
The primary drawback of this design is reduced incident light angle response compared to a non-microlens design
and Dalsa's that
At low f-numbers, microlensed pixels can suffer from vignetting, pixel crosstalk, light scattering, diffraction and reduced  MTF ...
The tradeoffs involved with microlenses are more pronounced with CMOS imagers since the microlenses are farther from the photosensitive surface of the pixel due to the “optical stack” of transistors.
« Last Edit: November 11, 2010, 11:39:31 AM by BJL » Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3418


« Reply #126 on: November 11, 2010, 02:21:51 PM »
ReplyReply

The Kodak sensors compared are both full frame CCD's, so as far as I can tell they have no signal buffer, just a little space at the edges for lateral overflow drains ... and one of the main claims in that Kodak paper is that the design reduces the space needed for the LODs, giving an impressively high 69% of the photosite being electron well (some is then masked off from direct illumination, but can still hold electrons). This loss to LODs is less than with the CMOS or interline CCD designs in most SLRs.

To repeat, the need for a smaller "window" (more masking) over the well in the microlens design is forced by the difficulties like pixel cross-talk that are made worse by microlenses; it is not some dumb stupid thing that Kodak did with the KAF-31000, degrading its performance in one respect, while knowing how to to better in another (the KAF-39000 without microlenses).

From the KAF-31600 data sheet, page 4, you linked to:
DESCRIPTION
The KAF-31600 is a dual output, high performance color
array CCD (charge coupled device) image sensor with
[...]
Microlenses are added for improved sensitivity.
The photoactive pixels are surrounded by a border
of buffer and light-shielded pixels
.

I interpreted that (enforced by SEM images of the sensel structure I saw somewhere) as all individual photoactive pixels being surrounded by other pixels. Obviously that would reduce the photosensitive area, and microlenses would be needed to refocus the light onto the sensitive areas. On second reading it could also mean (and would be more likely) that there is a row of shielded pixels around the entire array of photosensitive sensels.
On the KAF-39000 sheet description, there is no mention of surrounding pixels/shielding, suggesting a significantly different design, at least that's how I interpreted it.

All I'm saying is that differences in design make comparisons difficult, and certainly not a basis to generalize on, and the (also mentioned in the datasheet, offset) microlenses were instrumental in increasing the overall sensitivity of the (more special ?) design. Despite the identical 6.8 micron sensel pitch, the datasheets mention different sizes for the sensor array, but perhaps they are measuring 2 different things, that also gave me the impression that there is something else going on. Maybe it's just inconsistent measurements, e.g. total area versus effective area. They also mention different numbers of pixels, confusing ...

I would rather draw general conclusions about microlenses based on identical sensor array designs, only differing in the use of microlenses or not, that's all. That will probably be data that's hard to find.

Quote
Since I am the only one providing actual data about actual sensors and quotes from sensor making companies, can you provide authoritative evidence to the contrary, like spec's a sensor with microlenses than has off-perpendicular sensitivity "at the center of the sensor array" as good as in that non-microlens sensor? You might find one such, a new one from Dalsa using its new "low profile" microlens design, but that approach only works with CCDs, not CMOS, as Dalsa itself explains, so I doubt that any recent mainstream SLR sensor (all CMOS with microlenses) has such performance.

We'll see what can be found, but not all manufacturers have such easily accessable datasheets as Kodak.

Quote
Until you come up with superior authority or arguments, I have to go with Kodak's statement that
the KODAK KAF-39000 Image Sensor (39 Mp) is designed without microlenses to maximize incident light-angle response ... the critical crosstalk angle is increased
and of the KAF-31000 that
The primary drawback of this design is reduced incident light angle response compared to a non-microlens design


You make it sound like we're in some sort of pissing contest, which we're not (I'm not). We agree for the most part, we just cannot find independent sources that publicly quantify what we'd like to know without additional guesswork.

Quote
and Dalsa's that
At low f-numbers, microlensed pixels can suffer from vignetting, pixel crosstalk, light scattering, diffraction and reduced  MTF ...
The tradeoffs involved with microlenses are more pronounced with CMOS imagers since the microlenses are farther from the photosensitive surface of the pixel due to the “optical stack” of transistors.

Yes, that's all true in general, "microlensed pixels" can suffer from all that but do so in differing degrees, and changing the fill-factor with microlenses does indeed change the MTF. No discussion needed. Now to find 2 comparable sensor array designs, preferably the same, where one has microlenses and the other one has not, and the same between CCD and CMOS (although we already know that CMOS has more layers in general, and thus sensitivity to tunneling effects). Then we will be able to quantify the differences, without design variables that make such a comparison shaky at best.

Cheers,
Bart
Logged
nass
Newbie
*
Offline Offline

Posts: 12


« Reply #127 on: November 11, 2010, 03:10:37 PM »
ReplyReply

Could I maybe ask what maybe a really dumb question? F stop is just length/width, ie a ratio. A relative measure, an indicator of lens mechanics, not an absolute measure of light. Something from history which has suited us well and is still used because, well, it is easily understood. You could have a lens with a piece of cardboard in the middle and it'd still be f 1.2 Smiley. Isn't the real problem here the uninformed belief that f stop is an absolute measure of light and therefore every f1.2 is the same throughout history, rather than just a calculation useful for that particular lens only?
Logged
pegelli
Sr. Member
****
Offline Offline

Posts: 581



WWW
« Reply #128 on: November 11, 2010, 03:27:04 PM »
ReplyReply

Could I maybe ask what maybe a really dumb question? F stop is just length/width, ie a ratio. A relative measure, an indicator of lens mechanics, not an absolute measure of light. Something from history which has suited us well and is still used because, well, it is easily understood. You could have a lens with a piece of cardboard in the middle and it'd still be f 1.2 Smiley. Isn't the real problem here the uninformed belief that f stop is an absolute measure of light and therefore every f1.2 is the same throughout history, rather than just a calculation useful for that particular lens only?

Hey nass, there's no dumb questions, only dumb (or wrong) answers  Wink. Hope mine doesn't fall in that category  Shocked

Question at hand here in this thread is not so much wether a 1.2 lens meets the definition you state and lets through all the light it's supposed to do, but wether the digital sensor behind it can capture and record this light. Mark Dubovoy's article talks about two effects. First a general darkening due to the rays coming from the outer rims of the lens wide open not being recorded and secondly that due to the same effect on the sensor the depth of field is different, as it might cut out preferentially the outer rims of the circle of confusion that is created when an object is out of focus. Mark claims the general darkening is compensated by a "secret" ISO boost that is happening in the body and if you look at Pierre Vandevenne's pictures there seems to be some evidence for that. On the other hand you can see in my post that I have not been able to reproduce the depth of field/circle of confusion effect with my f 1.4 85 mm lens. Doesn't mean it doesn't exist for other lenses/sensors, but at least I'm "safe"  Grin
« Last Edit: November 11, 2010, 03:28:42 PM by pegelli » Logged

pieter, aka pegelli
PierreVandevenne
Sr. Member
****
Offline Offline

Posts: 510


WWW
« Reply #129 on: November 11, 2010, 05:55:09 PM »
ReplyReply

nass said

Quote
Could I maybe ask what maybe a really dumb question? F stop is just length/width, ie a ratio. A relative measure, an indicator of lens mechanics, not an absolute measure of light. Something from history which has suited us well and is still used because, well, it is easily understood. You could have a lens with a piece of cardboard in the middle and it'd still be f 1.2 . Isn't the real problem here the uninformed belief that f stop is an absolute measure of light and therefore every f1.2 is the same throughout history, rather than just a calculation useful for that particular lens only?

It's not a dumb question. That type of lens exists, for example in catadioptric telescopes, with a central obstruction. The key point (as you suggest) is that they would not behave as expected for the photographer who uses those ratio to reason about the proper exposure. In very, very broad terms (because each subpoint of that topic could generate longish discussions)

- their geometry would remain F/D 1.2
- their resolving power would remain roughly equivalent to an unobstructed lens (at the extreme, that is the principle behind large base interferometry and interferometry in general)
- their contrast would be "modified" - this is really a complex issue see http://www.astrosurf.com/legault/obstruction.html for more details
- their light gathering ability would be reduced by the obstruction (which is probably the main reason why they wouldn't behave as the F/D 1.2 photographers expect)

And I don't even dream thinking about the analysis of marginal rays in micro-lensed sensors ;-)
Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3418


« Reply #130 on: November 11, 2010, 06:37:07 PM »
ReplyReply

Mark claims the general darkening is compensated by a "secret" ISO boost that is happening in the body and if you look at Pierre Vandevenne's pictures there seems to be some evidence for that.

Hi Pieter,

Indeed, I have been able to detect that as well, but the weather (and the IRIS software) hasn't been nice enough to do a quantitative test yet. Sofar it seems like a fraction of 1/3rd of a stop gain, but the more accurate test will tell it more accurately (for my 1Ds3+ 2 lenses).

Quote
On the other hand you can see in my post that I have not been able to reproduce the depth of field/circle of confusion effect with my f 1.4 85 mm lens. Doesn't mean it doesn't exist for other lenses/sensors, but at least I'm "safe"  Grin

I haven't been able to detect that either, on a 85mm f/1.2 lens. The OOF blur 'disks' follow a perfectly predictable path, with very high reliability (R^2=0.9985) in a power function regresssion test. I'll try and post some results tomorrow.

Cheers,
Bart
Logged
pegelli
Sr. Member
****
Offline Offline

Posts: 581



WWW
« Reply #131 on: November 12, 2010, 01:08:47 AM »
ReplyReply

Hi Pieter,

Indeed, I have been able to detect that as well, but the weather (and the IRIS software) hasn't been nice enough to do a quantitative test yet. Sofar it seems like a fraction of 1/3rd of a stop gain, but the more accurate test will tell it more accurately (for my 1Ds3+ 2 lenses).
That would be nice, but to be fair I'm not to worried about this effect for my (amateur) shooting. Not all my exposures are spot on and I sometimes make larger corrections in my raw converter  and still get acceptable results. Yes pixel peeping the noise increase, but for an A3 print or downsized web presentation I find changes of up to a 1/2 stop are not significant. However I understand that very serious amateurs/pro's who want to get the absolute maximum quality out of their exposures even less than 1/3 stop can be a big deal.

I haven't been able to detect that either, on a 85mm f/1.2 lens. The OOF blur 'disks' follow a perfectly predictable path, with very high reliability (R^2=0.9985) in a power function regresssion test. I'll try and post some results tomorrow.
LOL, I only tested f 2.8 vs. f1.4 so I guess my R^2 is 1.000 (if you want to find a straight line relationship only do 2 experiments Grin), but you seem to be way more rigorous and scientific which I applaud. I found the DOF story a bit far fetched from the beginning and I'm glad we're starting to see the evidence this effect can be dismissed.

« Last Edit: November 12, 2010, 01:10:51 AM by pegelli » Logged

pieter, aka pegelli
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3418


« Reply #132 on: November 12, 2010, 06:07:32 AM »
ReplyReply

I found the DOF story a bit far fetched from the beginning and I'm glad we're starting to see the evidence this effect can be dismissed.

Well, there's not much wrong with a hypothesis being rejected, as long as one gains insight in the proces it still counts as progress Wink

Here's a quick summary of my findings with regards to the sensel tunnel shading effect on DOF.

I took a series of images at 1/3rd stop intervals, keeping the exposure time constant. The subject was a MagLite 'Solitaire' (very small, single AAA battery powered). I focused the lens at infinity, and positioned the tiny bare bulb (just unscrew the front of the torch) at a little more than 1 metre distance in the approximate center of the image. That should create a nice out-of-focus (OOF) highlight. Off-center measurements will suffer from vignetting and light fall off, so I stuck to center image testing.

The images were evaluated in Photoshop with the ruler tool, measuring the approximate (because the 8 segment aperture is not exactly circular) diameter of the OOF highlight images. I tried to measure at the same gradient of darkening at the edge of the OOF highlight 'circle'.

I tabulated the measurements in a spreadsheet, and converted the numbers of pixels across to millimetres by multiplying with the sensel pitch of 6.4 micron. I converted the diameters to approximate area (pi * (diameter/2)^2). The rounded F-numbers were replaced by their more exact mathematical equivalents, and a plot was made in a graph, and a regression trendline was calculated. The regression trendline was compared to the actual observations in the table, and an overall very good fit was found.

The trendline does not follow a perfect power of 2 relation, although it is pretty close. Afterall we are talking about a mechanical aperture and I know I have some issues with mine. The main point is that there is no evidence of a suddenly degrading fit at the widest apertures. On the contrary, the trend is very stable with a low average deviation, and the widest apertures show no sign of a reduced diameter.

Based on this, I tend to rejecting the hypothesis that the DOF at the widest apertures is significantly impacted by sensel tunnel shading.

Cheers,
Bart
« Last Edit: November 12, 2010, 06:23:35 AM by BartvanderWolf » Logged
sandymc
Full Member
***
Offline Offline

Posts: 248


« Reply #133 on: November 12, 2010, 06:25:34 AM »
ReplyReply

The photoactive pixels are surrounded by a border
of buffer and light-shielded pixels
.

I interpreted that (enforced by SEM images of the sensel structure I saw somewhere) as all individual photoactive pixels being surrounded by other pixels. Obviously that would reduce the photosensitive area, and microlenses would be needed to refocus the light onto the sensitive areas. On second reading it could also mean (and would be more likely) that there is a row of shielded pixels around the entire array of photosensitive sensels.
On the KAF-39000 sheet description, there is no mention of surrounding pixels/shielding, suggesting a significantly different design, at least that's how I interpreted it.

Pretty much all sensors have have an band of masked pixels around the active array; these are used to measure and compensate for dark current, the amount of charge that accumulates in a pixel even if no light falls on it. Usually this band is of the order of 10-20 pixels wide.

No sensor that I know of has any masked pixels in the active array.

As regards microlenses, CMOS and CCD sensors are different as regards their need for microlenses. Generally, a CCD array is just a big array of active cells, so the ratio of active pixel area to "dead" space, the "fill factor" is high. Also, CCD cells tend to be relatively shallow. CMOS cells need a lot of extra circuitry - amplifiers, etc - round each pixel, so the fill factor is lower, down to as low as 20-30%, and the pixels tend to be deeper. The fill factor is the reason for the so called "back lit" sensors. So when Kodak talk about not using microlenses to maximize sensitivity, be aware that's for a CCD sensor with a relatively high fill factor. A CMOS sensor is a very different proposition.

Sandy
Logged
PierreVandevenne
Sr. Member
****
Offline Offline

Posts: 510


WWW
« Reply #134 on: November 12, 2010, 06:28:26 AM »
ReplyReply

When I got my 1.2 lenses, I spent a few hours obliquely shooting calipers and rulers at different aperture settings. The progressive reduction in DOF was quite obvious (I haven't kept the images, a few GB of raw images of calipers and rulers just add to the size of the backups). Of course, I can't be sure it wouldn't be even better with film. That's a test that's a bit hard to make in controlled conditions.

Also, I did notice that portraits shot at 1.2 mandated a plane for the eyes that was perfectly parallel to the sensor, whereas 1.6 offered a safety margin.

But I have to say I was a bit surprised by the effect of the gain boost, which I expected to be measurable but not visible. There's also something I noticed and that I will try to investigate a bit later: the images shot at 1.2 and 1.6 seem roughly identically exposed, with the 1.2 gain boost and, somewhat surprisingly the same shutter speed (1/1250) in that case.

Since the conditions weren't really stable (I may have been casting a shadow, the economic bulb could have still been heating up...) I can't be sure of what it means, but keep an eye for this in more controlled tests
Logged
BJL
Sr. Member
****
Offline Offline

Posts: 5120


« Reply #135 on: November 12, 2010, 11:24:28 AM »
ReplyReply

First, let me note that I am not offering precise quantification of the magnitude of the effect we are discussing; after all, the data id not for any of the Canon, Sony or Nikon (CMOS or ILT CCD) sensors that the original article is dealing with; just with what seems an clear fact, that microlenses cause a significant fall-off in sensitivity as light comes from an off-perpendicular angle, including the light form the outer parts of there light cone from a low f-stop lens (like f/1.2 and maybe f/1.4) even at the center of the frame. But for reasons I have stated before and restate below, it is likely if anything to be greater with current DSLR front-illuminated active pixel CMOS sensors than with Kodak's FF CCDs.

Second, even if one denies that microlenses are the cause, the effect is still there in every single microlenses sensor I have data for (four different FF CCDs of three different pixel spitches from Kodak.)

... On second reading it could also mean (and would be more likely) that there is a row of shielded pixels around the entire array of photosensitive sensels.
The second reading is correct: this is a border around the edge of the entire sensor, not within each photosite. This is made clear in the early pages of the long spec document for the KAF-39000, and in every Kodak FF CCD long spec: see http://www.kodak.com:80/global/en/business/ISS/Products/Fullframe/
Quote
Despite the identical 6.8 micron sensel pitch, the datasheets mention different sizes for the sensor array, ... They also mention different numbers of pixels, confusing ...
It is not at all confusing if you read a bit more carefully: these are sensors using the same basic 6.8 micron photosite design but of different sizes (44x33mm vs 49x37mm) and thus of different pixel counts (39MP vs 31MP). But that difference is irrelevant to per pixel performance characteristics.
Quote
All I'm saying is that differences in design make comparisons difficult, and certainly not a basis to generalize on, and the (also mentioned in the datasheet, offset) microlenses were instrumental in increasing the overall sensitivity of the (more special ?) design.

I would rather draw general conclusions about microlenses based on identical sensor array designs, only differing in the use of microlenses or not, that's all. That will probably be data that's hard to find.
The difference in overall sensor size and pixel count is irrelevant to the question we are discussing: we are looking at what happens at an individual photosite, anywhere on the sensor, when light strikes it at various angles, and Kodak gives data for those two contemporary sensor designs for what happens at a photosite at the center of the sensor. By the way, here is another earlier one with microlenses but not offset, that KAF-8300 of the Olympus E-300 and E500: http://www.kodak.com:80/global/plugins/acrobat/en/business/ISS/datasheet/fullframe/KAF-8300LongSpec.pdf
and the KAF-18000, an 18MP, 44x33mm sensor with 9 micron pixels and non-offset microlenses:
http://www.kodak.com:80/global/plugins/acrobat/en/business/ISS/datasheet/fullframe/KAF-18000LongSpec.pdf
which you might want to copare to the same generatio 22MP, 49x37mm KAF-22000 with no microlenses and far less off-perpendicular fall-off:
http://www.kodak.com/global/plugins/acrobat/en/business/ISS/datasheet/fullframe/KAF-22000LongSpec.pdf
The angular fall-off is even worse for the older KAF-5101 of the Olympus E-1, with 6.8 micron pixel pitch and non-offset microlenses, but that is no longer at Kodak's site.
It becomes hard to conclude anything that than, consistently over eight years of putting microlenses on FF CCDs, Kodak has been forced to sacrifice off-perpendicular sensitivity fall-off as part of the price of adding microlenses.

The size difference is however relevant in a different sense: despite the substantial advantages in QE that microlenses give when Kodak uses them in FF sensors (about doubling QE and thus adding one stop of sensitivity) and despite Kodak having had this technology since the KAF-5101 from 2002 with this same 6.8 micron pixel spacing, Kodak has never used microlenses on its larger MF sensors, the ones about 49x37mm, while using them on three generations of its smaller 44x33mm FF CCDs for MF (KAF-18000, KAF-31600, KAF-40000). The reason seems obvious, especially since Kodak more or less states it in text I have already quoted: adding microlenses has the unfortunate side-effect of causing greater fall-off in sensitivity to off-perpendicular incident light, causing a kind of vignetting, and this increases as distance from the center of the frame increases. The effect is thus more tolerable with the smaller radius (corner distance) of the 44x33mm MF sensors than with the larger  radius of the 49x37mm sensors. Before Kodak had off-set microlenses, it did not use microlenses on any MF sensor.

I ask again: can anyone suggest any plausible reason why Kodak would hamper every one of its microlensed FF CCD sensors in this way if it could combine the QE advantages of microlenses with the greater off-perpendicular sensitivity that all its non micro-lensed FF CCD sensors have, going back many years?

Or to put it another way, why does Kodak continue to hamper the sensitivity of all its largest 49x37mm MF sensors with substantially lower QE by not offering any of them with microlenses, except due to its inability to add microlenses while avoiding the vignetting problem due to sensitivity that declines too fast as the angle of incidence increases?

And to the idea that this is just inability on Kodak's part, I will note again that (a) Dalsa says the same thing about off-perpendicular sensitivity fall-off being a disadvantage of microlenses, and (b) as Dalsa explains, the problem is likely to be worse with all recent DSLR sensors because they are all front-illuminated CMOS sensors, and thus microlenses need to be further from the wells.
Logged
BernardLanguillier
Sr. Member
****
Offline Offline

Posts: 7749



WWW
« Reply #136 on: December 13, 2010, 04:45:41 AM »
ReplyReply

In case someone is still interested in this, the latest edition of Chasseur d'Image has an article in French that I find to be a more accurate explanation of the issue Mark wrote about.

Regards,
Bernard
Logged

A few images online here!
pegelli
Sr. Member
****
Offline Offline

Posts: 581



WWW
« Reply #137 on: December 13, 2010, 07:32:39 AM »
ReplyReply

I'd be most interested in the DoF change suspicion. There's posts in this thread that show this effect could not be demonstrated with f 1.4 and f 1.2 lenses on two different bodies but maybe Mark and/or DxO have investigated this further and can provide some new insights or measurements.

I think the discussion (and subsequent articles) have the artificial ISO gain pretty well understood (despite the lack of response from the manufacturers), but it's been very silent on the DoF issue where I have more doubts about.
Logged

pieter, aka pegelli
Herb19
Newbie
*
Offline Offline

Posts: 14



WWW
« Reply #138 on: December 22, 2010, 04:12:51 AM »
ReplyReply

Lens/camera designers from Fuji are very aware of the subject raised by Mark. They do not seem to follow the road of secretly boosting ISO.

A quote from: http://www.finepix-x100.com/story/


2Road to the F2 aperture value.

    * Designing an F1.6 or F1.8 lens is not so difficult; however, in the case of a digital camera, even if an aperture larger than F2 is used, the light receiving elements on the sensor cannot effectively use the brighter portion of the incoming light because of low incident light gathering efficiency.
« Last Edit: December 22, 2010, 04:20:17 AM by Herb19 » Logged
PierreVandevenne
Sr. Member
****
Offline Offline

Posts: 510


WWW
« Reply #139 on: December 22, 2010, 08:01:30 AM »
ReplyReply

I did a few tests

http://www.datarescue.com/photorescue/freefiles/compare40d5dmkII.jpg

http://www.datarescue.com/photorescue/freefiles/5dmk2.jpg

http://www.datarescue.com/photorescue/freefiles/40d.jpg

40D / 5DMK2 - tripod - remote trigger - 1 meter from target - auto exposure - lighting poor on purpose to better see the noise. Shot in RAW, converted to jpeg at max quality with no adjustment whatsoever, ISO 800

AFAIC

- the gain boost is quite visible, maybe a bit less on the 40D, but the auto-exposure chose different parameters.
- the practical impact, at least at pixel peeping level, of the gain boost is more significant than the raw noise number indicate, probably because noisy pixels have a bigger than expected impact on the RAW conversion.
- the DOF decreases as expected (it is very visible on near tangent images of printed text, as posted earlier) at least on the 5D
- the 40D doesn't seem to exploit that DOF increase fully, but that's just a gut feeling, and DOF is bigger anyway.

Logged
Pages: « 1 ... 5 6 [7] 8 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad