dynamic range of ccd

[Mike Brennan]

Recently we were discussing dynamic range of ccds and 10 bit vs 12 bit A-Ds. Came across this article from Roper Scientific. Seems that there *is* a conection between the type of ccd and bits assigned to A-D conversion.

Roper are mostly refering to cooled CCDs where noise is very low.

They make the point that commercial ccd applications use 16bit A-D to maximise dynamic range of a particular type of ccd.

I have been trying to find out why Sony had remapped their A-D with the f900 upgrade.

Could it be that there is more dynamic range available in the ccds than 12 bit can deliver? By remapping have they have created a kind of log response? Wouldn't it be interesting to be able to manipulate the A-D conversion itself rather than its output.

 

Mike Brennan

 

 

http://www.roperscientific.com/library_enc_dynamic.shtml

 

 

"As a specific example, consider a Kodak 1401E CCD, which has a full well capacity of 45,000 electrons. At a typical readout rate of 1 MHz, the read noise is 11 e-. The dynamic range of this chip is therefore 45,000:11, or 4,091:1. In order to take full advantage of this dynamic range, cameras incorporating Kodak 1401E chips usually utilize a 12-bit A/D converter (4096 gray levels). It is important that the camera's readout and signal-processing electronics be optimized so that low read noise is maintained, otherwise the dynamic range will be compromised.

 

To extend dynamic range beyond the 12 bits given in the previous example, a camera with a lower read noise or a CCD with a larger full well capacity is required. Full well capacity is related to pixel size. For instance, the Thomson 7895 has a capacity of 375,000 e- and a read noise of 5 e- rms at 40 kHz. The dynamic range is thus 75,000:1. In commercial cameras this is usually coupled to a 16-bit A/D converter (65,536 gray levels)"

[Guest Pete Wright]

Interesting information. I have some questons:

 

Is 12 bit equal to 12 F stops and 16 bit equivalent to 16 F stops?

 

What is the dynamic range of negative film and of reversal (projection) print?

 

Are lattitude and dynamic range the same thing?

 

A comment: 16 bit is nice but how do you work with it? Most NLE's are 8 and 10 bit. Are there any 12+ bit ones?

 

Pete

[Phil Rhodes]

Hi,

 

This is all correct, but I stand by my assertation that you can't increase the dynamic range of a CCD by changing the ADC convertors. Obviously, you need to pick the right convertor for the CCD.

 

Phil

[Filip Plesha]

The 16-bit A/D converters are just a marketing trick..

 

A/D converters are getting cheaper all the time and a lot of manufacturers

use more and more just to trick the cunsumers. This usually goes for the consumer market, but I think a lot of pro manufacturers use it too...

 

for example there are a lot of scanners for still images out there

that use 16 bit a/d converters. THere is no way on earth that they can fill

that potential. I mean a half-million dollar cine scanner is using 14-bit

converter and a 20 000 dollar Imacon is using 16-bit. This is absurd.

The difference is that there is no need for a lot of advertising for motion picture scanners, so the manufacturers can be honest and put a converter than

does serve a purpuse. (but even that is under a questionmark)

[harryprayiv]

The 16-bit A/D converters are just a marketing trick..

 

A/D converters are getting cheaper all the time and a lot of manufacturers

use more and more just to trick the cunsumers. This usually goes for the consumer market, but I think a lot of pro manufacturers use it too...

 

for example there are a lot of scanners for still images out there

that use 16 bit a/d converters. THere is no way on earth that they can fill

that potential. I mean a half-million dollar cine scanner is using 14-bit

converter and a 20 000 dollar Imacon is using 16-bit. This is absurd.

The difference is that there is no need for a lot of advertising for motion picture scanners, so the manufacturers can be honest and put a converter than

does serve a purpuse. (but even that is under a questionmark)

So you're saying that the Dynamic Range of the CCD sensors hasn't yet reached the point where the entire 16bit range is filled (because even the best sensors can get away with 12bits instead of 16bit), so companies who are marketing a camera with a chip with less dynamic range than their competitor's more expensive chips (that probably have higher dynamic ranges) are really doing it even though it might not be necessary for their specific cameras? I find that odd...but perhaps true.

[Filip Plesha]

This is a well known fact in the bussines of digital still cameras (pro and consumer) and still image CCD scanners (pro and consumer).

For example there are VERY FEW CCD scanners on the market that can

scan a slide without loosing the dynamic range.

Yet if you read the specs, EVERY better CCD scanner that you can buy claims to be albe to capture the full dynamic range of slides. The experience shows

that the first statement is true. There is a reason why drum scanners are used for slides. But they all use 16-bit converters. If they could all fill that space, nobody whould bother to use drum scanners.

So yes, this is true sadly, but I am not generalising.

I am not saying that CCD technology can't give you real 14-bit images.

I am just saying that a lot of medium qualitty CCD's are claimed to do so.

 

I am sure that there are sensors that can fill 14-bit space

but they are probably used in things like cineon scanners.

[Luke Prendergast]

You don't quite understand what bit depth entails, or perhaps what dynamic range means. Phill addressed this in another topic I think.

 

1 bit or 24 or 1024 are not definitions of dynamic range. It is the number of gradations in between that is affected by bit depth. 1 bit is either on or off, 1 or 0, black or white. 16 bits is 256 levels or gradations, but the end stops are still black and white. Noise is the limiting factor, when the gradations are finer than the noise in the signal, but that's another topic.

 

Your 24 bit DVD audio isn't inherently louder than your 16 bit CD player is it? Silence isn't any quieter with 24 bits.

[Phil Rhodes]

Hi,

 

Well silence might actually be noisier with more bits, since the toggle space below 1 LSB is lower!

 

Phil

[Filip Plesha]

mr. Smith..

 

You are only half right..

 

yes, of course that the number of bits mean the number of gradations, but

it allso allows more dynamic range.

 

8-bit images are not capable of 10 stops of dynamic range simply because

there is no room to double up the value 10 times.

 

24-bit audio will not give you a louder sound. But it will give you more dynamic range. You can always make both 16 bit and 24 bit sound equally loud by

turning the volume up, but the difference is that with 16-bit you won't be able

to play softer and loud at the same volume setting.

 

Remember how DVD's can have a great dynamic range, they can play a whisper

and a thunder strike in the same recording with the same volume setting.

CDs flatten out the difference.

 

16-bit and even 32-bit (per chanell!!) file formats like EpenEXR are not used

because of the number of gradations, but because of the extended dynamic range.

With such formats you can for example save a bitmap image that will have recorded image information from an daylight interior and and exterior outside the window in the same frame, all you have to do is use a LUT you want.

You can ,for example, create one lut for viewing the outside details and another for interior detail. Both resulting images will be high qualitty.

Even film can't record such range.

try to do the same with a 8-bit image.

 

And what I have said about the manufacturers is true for both the dynamic range and the number of gradations as you mention.

[Luke Prendergast]

By your argument there can be how many stops difference between black and white in a 1 bit system? or a 16 bit system? Black and white are not more black and more white in 48 bits than in 1 bit.

 

The luminance of black and of white is not changing. 0 equals minimum level equals black the same as 00000000 does. 1 equals maximum level equals white as does 11111111.

 

I should maybe point out that I am a professional in the field of electronics.

[Rob Belics]

I used to be an expert in electronics but have forgotten a lot. What I've forgotten is the terminology for what is being discussed here. "Dynamic range" is the "range of change" but "resolution" is the definition for the number of steps in an a/d converter.

 

Noise plays a factor, somewhere in all this, because noise can occupy some of the lower and higher level bits so your effective range is lower.

 

8-bits, by the way, is 256 levels, not 16 which is 16384.

[Phil Rhodes]

Hi,

 

> 8-bit images are not capable of 10 stops of dynamic range simply because

> there is no room to double up the value 10 times.

 

That's not the point - Mr. Prendergast is right. You're confusing resolution and dynamic range, which is probably what the camera manufacturers want you to do!

 

Phil

[Filip Plesha]

mr. smirth, you are missing the point..

 

you are talking about displayed white and displayed black..in that case both 8-bit and 32-bit are the same, but this is only after you apply a LUT to them.

 

Remember than in 10-bit cineon images the white point is not the maximal value.

What you see on screen as white is not 1023 but 685. Everything pass 685 is

equal to white on screen, but in the file itself there are details up there,

and you can "pan" your little "black-white window" and choose do you want values

from say 25- whatever or from 500 to 1024 or whatever.

This is where the extended dynamic range is, you don't see it when you apply

the LUT, but it gives you the oportunity to use more "exposure settings" for

one file.

 

And I am not talking about cameras, I am talking about computer digital files, perhapse this is where our disagreement lies. Cameras always record everything in display contrast, there is no need for LUT's to view the image.

[Phil Rhodes]

Hi,

 

> you are talking about displayed white and displayed black..in that case both 8

> bit and 32-bit are the same, but this is only after you apply a LUT to them.

 

Again that's nothing to do with bit-depth. Code value 685 is considered white on systems which render the natural logarithm of the input data and there is noticeable room above it for things like direct light (looking into the sun) and specular hilights. It has no more dynamic range than an equivalent 8-bit image; I can downsample a 10-bit log DPX into an 8-bit linear one and it doesn't chop the hilights off. To prove this, download this file and turn it into an 8-bit TIFF using a utility which understands log data:

 

www.kodak.com/US/en/motion/support/dlad/dlad_2048X1556.cin

 

> I am talking about computer digital files

 

It doesn't make any difference. There is still no absolute relationship between bit depth, or even linearity, and dynamic range.

 

> Cameras always record everything in display contrast, there is no need for

> LUT's to view the image.

 

This is absolutely not true; something like the Viper records a very strange-looking flat image if you choose to do no correction.

 

Much as I hate to make an appeal to authority, I have spent the last two weeks working for Filmlight on their Baselight software (Tomb Raider 2, Cold Mountain et. al.) and deal with these issues on a daily basis.

 

Phil

[Mike Brennan]

"have spent the last two weeks working for Filmlight on their Baselight software (Tomb Raider 2, Cold Mountain et. al.) and deal with these issues on a daily"

 

I don't disagree with anything anyone has said however reliable info about what happens with A/D in cameras is hard to identify. Plenty of post experts who know their stuff.

Roper make sensors, their article may be just vauge. Maybe there is something in A/D hardware that makes this true?

This topic started with the concept of mapping A/D to control compress highlight end of ccd response (something Sony say they have done on Mark 3 f900 by remapping.) Would be good to get laymans explaination.

 

Roper's article says that larger dynamic range of larger size pixels on cooled ccd sensors needs or benifits from 16bit A/D. Maybe they are refering to the overall greyscale representation/capture?

 

 

Guess we'll ask Roper on Monday.

 

Anyone know who makes the Genesis CCD?

 

Mike Brennan

[Filip Plesha]

Ok Phil, your arguments are ok, and you have me confused now..

 

Where is the advantage of the dynamic range then? If it is not in the files,

I don't udesrand where is it then.

Everyone keeps talking about the connection of bit deph and the dynamic range.

Not only that, but ILM developed their 16-bit and 32-bit Open EXR to have

the extended dynamic range.

 

And these 32-bit images are usually called HDR images "high dynamic range"

 

I don't understand...

[Rob Belics]

I may have missed a point here but if the range of available signal is, say, from 0 to 1 and you use an 8-bit converter, "stepping" between values is more noticeable than if you had a 12-bit converter. 16 and 32 will show less of this stepping.

 

In HDR, you are translating and moving dynamic range values into a smaller area. If you took an outdoor scene that has a range of zero to five but wanted to display it on a computer screen that has a range of zero to one, it's easier to fit the high bits into that zero to one range if you had more bits to work with, hence the value of using 16 and 32 bits.

 

It creates a smoother, less noticeable fit.

[Filip Plesha]

Yes Bob, i think you are right.

This may be the missing link between what Smith and Phil were saying

and what I was saying (only i was missing the main point)

 

It is possible to use HDR in 8-bits but if you take your display

range to be from say 20 to 100 it will show image degradations because

of lack of steps. That is probably why 16-bit is used for high dynamic range

because the qualitty is still good when you "strech" a section of the dynamic range

into display contrast.

 

This could be it

[Phil Rhodes]

Hi,

 

Yes, that's it. And there's only one T in "quality."

 

Phil

[Filip Plesha]

I'm sorry, English is not my native language..

i have been writting it with two t's for a long time, i didn't know that

[Rob Belics]

That's another entry to another thread. :P

[Phil Rhodes]

Hi,

 

Ah well apologies then - your English was otherwise so good I assumed you were a native speaker of the language.

 

Same with "writing," by the way. :)

 

Phil

[Filip Plesha]

I am from Croatia.

If you have ever seen Rade ?erbed?ija (Serbedzia or whatever it is spelled in US) act (Stigmata, mission imposible 2, mighty joe young, eyes wide shut etc.) ,you get the picture how Croatians speak english (he is allso from Croatia) :) It is usually horrible, i mean the pronunciation even though he is one of the worse cases. That's the reason he is usually cast to play a Russian.

[Max Jacoby]

If you have ever seen Rade S<caron>erbedz<caron>ija (Serbedzia or whatever it is spelled in US) act (Stigmata, mission imposible 2, mighty joe young, eyes wide shut etc.) ,you get the picture how Croatians speak english

"Doctor? You a medical Doctor? I have this problem with my hair. It falls too fast..."

[Filip Plesha]

Yea that's him.. :)

 

But , he is not avery good representative of how people look around here.

He looks like he has been living in a cave for a year or two. :)

But he is very respected around here, he has played in many Croatian (and Serbian too) films and he is considered as one of out finest actors.

 

Goran Visnjic is allso a Croatian actor, but he is a newcommer.

He is best know for his role of the "undead" Bulgarian in Practical magic,

and Dr. Luka in TV series "ER", other recent projects included the voice of

Soto in "Ice age" , "The deep end" etc.