Oversized CCD chips

[Barry Green]

<<<-- Originally posted by A. J. deLange :
What I'd really like to know is how the extra 33% (4/12) pixels in 16:9 mode get stuffed into the DV stream-->>>

They don't. DV is 720x480 regardless of whether it's meant to be interpreted as widescreen or as "normal" 4:3.

The CCD pixels do not necessarily bear a 1:1 relationship to the DV frame pixels. Obviously, in the case of a 960x480 CCD and a 720x480 DV frame, they cannot bear a 1:1 relationship. The CCD is an analog device, and in the process of sampling that analog device to convert it to a digital signal, the 960x480 pixels will be sampled down to 720x480.

[Boyd Ostroff]

>> posted by Barry Green : They don't. DV is 720x480 <<

A.J. - Barry is correct. Have a look at this thread for more discussion.

[A. J. deLange]

I took a vernier caliper to my XL1s this evening and while I can hardly claim a lot of confidence in the measurement (because one has to eyeball from the blades of the caliper to the superposed images of the chips which are quite distant from the caliper) I got 0.221" for the long dimension. This is 5.6 mm which means that the chip is probably a 1/3 incher with long dimension 4.8 mm (according to the Edmund site referenced earlier). But if I stubborly stick to my measurement the chip dimensions would be 0.221 x .166 for a diagonal of .276". This would make the XL2 4:3 active area about 0.225.

Now onto the replies. First, thanks. What I am actually looking for is the details of the algorithms. What I meant by "stuffed" is that DV25 has a fixed bit rate which will support 720x480x3x30 pixels per second whereas the XL2 is gathering 960x480x3x30 which is higher so that either the the number of pixels must be reduced by subsampling or the bits per pixel must be reduced by a more efficient compression algorithm (which is going to have a cost in either spatial or temporal resolution or both). I suspect that the algorithm tries to emulate an anamorphic lens so that programs like FCP can work with the data. This would mean mapping 1 CCD pixel to a tape pixel at the center of the picture and more than 1 CCD pixel (horizontal direction) to an output pixel at the edges. This would mean that the horizontal resolution at the edges of the picture will be worse than at the center. As I imagine the actual algorithms are proprietary I'd settle for some TV lines or MTF data. Obviously this is the most meaningful (or at least most quantitative) way of evaluating the claims of improved picture quality in the XL2. The data I'd like to see are MTF of XL1s and XL2 in 4:3 mode and MTF at center and edge in 16:9 mode for the XL2 as compared to the XL1 with anamorphic lens.


Cheers, A.J.

[Boyd Ostroff]

You sort of lost me in all that tech talk :-) But I think it's just a linear mapping, not one that varies from the center to the edges. Are you suggesting that they would deliberately introduce distortion? Anamorphic DV is just like regular DV except the pixel aspect ratio is different. If you use my photoshop example in the other thread I referenced it will create a proper anamorphic image.

So somehow the camera reads 4 of its 960 horizontal pixels and outputs only 3 to the 720 horizontal pixels on tape. None of this is a new concept, there are other inexpensive camcorders that shoot native 16:9, such as the PDX-10 and GS-400 both of which work with a raw image that's 1152 pixels wide.

Sorry if I didn't understand what you were getting at. Not knowing anything at all about camera design and internal processing, I always assumed that the CCD data was read into some sort of buffer where the transformation is made using simple math. Seems like 1/30 second is quite long in terms of computer processing time.

[A. J. deLange]

Boyd,

I was indeed suggesting that the anamorphic distortion was introduced non linearly so that when taken out by the projection lens the center of the picture (the part you'd look at with the fovea) would have higher resolution than the sides which you would look at with the lower resolution part of your retina. But your comments and a little further research have me convinced that in 16:9 video cameras, at least, that's not the case. It apparantly is simply a linear squeeze with an anamorphic lens being composed of two cylindrical lenses with the vertically oriented (axis of the cylinder) one having a focal length shorter than the horizontally oriented one. In a 16:9 CCD the same effect is gotten by sample rate conversion in the horizontal dimension with 4 samples in and 3 out. Sample rate conversion by rational fractions (such as 3/4) can be done with relatively simple math. Whether you'd consider it simple in absolute terms is a different matter. Today's electronic devices do incredibly sophisticated processing in incredibly small devices (and don't forget that the Canon camera use pixel shift technolgy, something else I'd like to understand, to increase the apparent resolution beyond what the CCD pixel count implies). The downside of downsampling in this way is that 25% of the video bandwidh (or horizontal resolution if you prefer) of which the CCD is capable is thrown away. A pity but apparently DV25 just isn't capable of handling it.

Given that that's all there is to anamorphism (from lens or CCD) all FCP (or any other application) has to do is recognize that the pixels from a DV tape should be interpreted as being 1.333 times wider than they are high if the anamorphic bit (I have found that the DV standard defines such a bit) is set. Piece of cake and it all fits. Thanks for pointing me in the right direction.


Cheers, A.J.

[Luis Caffesse]

Okay, not that this really matters at all, but something occured to
me today about Canon's specs.

Most people assume that a 1/3" CCD is actually .33 inches in
diagonal.

So, it seems that Canon's specs are actually RELATIVE
measurements that assume a .33 inch diagonal on the chip.

I did the calculations the other day, and said that in 4:3 the
diagonal on the XL2 is actually closer to .17 inches, and in 16:9 it
is closer to .211 inches.

BUT, if you do the math, and assume that the chip is .33 inches in
diagonal (which we know that it isn't), then the diagonals for the
image area in 4:3 and 16:9 match Canon's specs.

IF the chip were .33 inches in diagonal, the XL2's image area
WOULD be .236 inches in 4:3 and .289 inches in 16:9

My point is, people assume a 1/3" chip is .33 inches in diagonal.
So maybe Canon figured it would be easier to give 'relative'
measurements of the image area than the actual measurements.

Imagine how confused MOST people would have been if Canon
published the actual measurements, and said that the image
area in 4:3 was roughly .17 inches. People would think that was
the same as a 1/6" chip.

Just a theory.

Ironic that the names for these chips came about to avoid
confusion, and now that tube cameras are long gone, these
names do nothing but confuse people.

-Luis


EDITED TO ADD:

Hey look at that, I'm a trustee

[Barry Goyette]

Wait a minute luis...this is where I was coming from the other day...saying there was a lot of assuming going on...and no-one was trusting the listed specs...

Anyway, as it stands now..I think the lens focal length proof...as detailed by Chris...is the most persuasive evidence that the effective areas are smaller than what Canon has posted.

Barry

[Luis Caffesse]

Barry,

Maybe I wasn't very clear in what I was trying to say.
I am not saying that their specs are correct.
The specs they have listed are impossible, as we have already discussed.

All I was saying was that it is now my theory that Canon's specs are
"relative" measurements, not actual measrements.

The ACTUAL measurements would be much smaller than what Canon listed
because of the fact that 1/3" CCDs are actually .236 inches in diameter.

BUT, if you assume that a 1/3" CCD is .33 inches (as most people do) then
you will get the same specs that Canon posted.
I did the math, and got the same numbers they posted for both 4:3 and 16:9.

My point was that they may have done that to avoid confusion, because
most people assume that 1/3" CCDs are actually 1/3" in diagaonal.

While their specs are still wrong, it is easier for them to list 'relative' measurments
than it is to explain the history behind CCD nomenclature.

I'm not saying this is a good idea, but I think that's what they may have done.
The point is, they may not have posted the wrong specs by mistake.
They may have posted the specs in relation to what they would be on a chip with a .33 inch diagonal, because that is what most people assume a 1/3" chip has.

Does that make sense?

-Luis

[Dan Vance]

Hmm....I dunno about that. The "names" for the formats, e.g., "1/3 inch," "1/2 inch," etc. are always expressed as fractions, and once you know the system, you know what those names represent.
Using a decimal number implies an actual measurement. Thus, the dimensions given on spec sheets for things like focal length, flange focal distance AND imaging areas are always in mm or decimal inch equivalents, never in fractions. I don't think you'll see any spec sheets that call a 1/3" chip a ".333" inch chip, because that has a totally different implication, and suggests that the diagonal image area is .333". But it may be that the person who wrote the spec in the sheet was a non-technical type who made a big (wrong) assumption.

[Chris Hurd]

<< But it may be that the person who wrote the spec in the sheet was a non-technical type who made a big (wrong) assumption. >>

My thoughts also. Most likely it was a marketing type who actually thought that a 1/3rd-inch CCD diagonal was 0.333"

[Luis Caffesse]

"once you know the system, you know what those names represent."

Sure, once you know the system.
The thing is, we're dealing with a prosumer camera, and most
users do not 'know the system.' You'd be surprised by some
of the people I've spoken to in the last week who, no matter
how I tried to explain it, continue to believe that 1/3" means
a chip is .33 inches. Then again, can you blame them?

But, whether or not it was done on purpose or by mistake
is kind of moot. If it was on purpose, it's rather misleading
to use decimals, you're right Dan. So chances are it was just
a mistake.

I guess my main point was that it is not simply a random typo,
(as I originally thought) the numbers would be valid IF the chip
were actually .33 inches in diagonal, which is what most people
think.

I'm assuming most of you have seen the posts, both on this
forum and other sites, where people are pointing to these
same specs to "prove" that in 4:3 mode the XL2 is using the
equivalent area of a 1/4" chip. They see the .236 inch diagonal
and that is what they assume.

Again, all of this is really unimportant.
I just thought it was interesting that the numbers were not
completely random, and made a "relative" amount of sense.


-Luis

[Dan Vance]

Yes, that really is intriguing how those numbers fit the scaling from .33." I wonder if we'll ever know how it really happened, and if so, if it will make sense!
I would think someone who is close to Canon (not mentioning any names...) could call someone in the tech department and get a straight answer and post it.

[Barry Green]

Yeah, getting into the math is not something I'm too interested in doing, to "prove" something one way or the other. Nobody really knows anything other than what the manufacturer tells us, right?

The proof is in the angle of view of the lenses. An XL1 delivers an angle of view consistent with that which is commonly called a 1/3" camera (regardless of how big the CCD actually proves to be, it's "called" 1/3"). The XL2, in 4:3 mode, delivers an angle of view consistent with that which is called a 1/4" camera. That's really all the proof you need...

[Jeff Donald]

The *Angle of View on lenses (obviously excluding zoom lenses) is fixed unless they have a back focus adjustment. Lenses produce a *Circle of Good Definition, this is the area that must cover the imaging device (CCD, CMOS, film etc.). The Circle of Good Definition (on a better quality lens) usually exceeds the radius of the imaging device by a millimeter or two. Poor quality lenses do not exceed the radius and hence Light Falloff or vignetting that occurs. Falloff can be improved by stopping down the lens. The general definition of allowable falloff is 50% (1 stop). Better quality lenses control falloff to 1/3 stop or less.



*Angle of View=The angle formed by the lines from the rear nodal point to the two opposite sides of the imaging device located one focal length from the rear nodal point, lens focused at infinity.


*Circle of Good Definition=The circular are in the image plane within which the lens forms an image having acceptable resolution.

[Luis Caffesse]

"An XL1 delivers an angle of view consistent with that which is
commonly called a 1/3" camera (regardless of how big the CCD
actually proves to be, it's "called" 1/3"). The XL2, in 4:3 mode,
delivers an angle of view consistent with that which is called a
1/4" camera. That's really all the proof you need..."


Actually, the math is somewhat important Barry.
Looking at the numbers, you'll see that the XL2 will deliver a field
of view which is somewhere between a 1/4" chip and a 1/3" chip
(albeit, it will be closer to a 1/4" chip).

So, it is not exactly right to say it is "consistent with that which
is called a 1/4" camera."

-Luis