Why Bayer

[Graeme Nattress]

Ronz, now you're asking things I have no answer to! I can't predict red future at all.

Sure, downsampling would make anything in the image that's undesirable smaller, but you'd still need as strong OLPF in relation to pixel pitch as you have now. If you only ever wanted to get 4k out of your hypothetical 8k sensor, you would use a stronger OLPF, and then you could just decimate or average.

Graeme

 

[conrad gaunt]

Ronz, now you're asking things I have no answer to! I can't predict red future at all.

Sure, downsampling would make anything in the image that's undesirable smaller, but you'd still need as strong OLPF in relation to pixel pitch as you have now. If you only ever wanted to get 4k out of your hypothetical 8k sensor, you would use a stronger OLPF, and then you could just decimate or average.

Graeme

I just thought you might have some inside knowledge on upgrade path. The NAB pdf still says the Red1 does 4k 1-60fps.
As for the hypothetical 8k sensor, in what way would the OLPF have to change then if pixel pitch doubled (which would surely give it twice the frequency response of a 4k sensor?), surely the frequencies you would need to remove with the LPF (ie, frequncies above the LP frequency) would be higher, by a factor of two roughly?

 

[Graeme Nattress]

Ronx, usually you base the OLPF on the pixel pitch, so if the pitch halves, you can halve the strength of the OLPF, all things being equal, within the bounds of actually making the OLPF, which is something I don't know enough about to go into any more detail on.

Graeme

 

[Barry Green]

The NAB pdf still says the Red1 does 4k 1-60fps.

The Red One sensor DOES do 4K from 1 to 60 fps.

You just can't record it. (at least, not yet)

Select "Open Gate" for your viewfinder, select 4K @ 30FPS, and what you're watching in your viewfinder is actually 4K at 60FPS. When in "open gate" mode, the sensor runs at twice the speed of the project's frame rate (meaning, 4K @ 24fps, the sensor is actually running at 48fps.)

Someday it may be able to record it, but right now you can see that the hardware does run at that speed. Don't know if that's enough to cover what the NAB PDF is claiming (and, hey, Build 16 isn't out yet -- maybe Build 16 includes 4K @ 60FPS?!)

 

[paulcurtis]

Graeme,

An interesting post and i totally understand the red approach and agree with it for the most part.

With a 3 CCD approach you can apply RGB gain to each individual CCD which is something you cannot do with a single sensor. So not only are the 3 x mono sensors themselves more sensitive but you can colour balance to a certain degree within the sensor hardware (tungsten/daylight etc,.).

CMOS sensors have hardware on each pixel, does this hardware, in Reds case, allow for individual gain being applied to a pixel? Or do you feel that this isn't a huge issue for single sensors?

Also you mentioned pixel pitch on 2/3rds vs Red but we don't know what the fill factor is on the mysterium (or do we?).

Going to 5k on epic seems like it could be a step in the wrong direction for sensitivity unless of course the fill factor or other hardware improvements have been made (which i kind of assume they have...), is it possible to comment about this?

thanks everyone for an interesting discussion, i can't wait to be shot down in flames

cheers
paul

 

[Radoslav Karapetkov]

Would it be fair to say that, if comparable to the RED One, the future Scarlet's 3K sensor will deliver:


- About 2,34 K Luma Resolution.


and


- About 1,71 K Chroma Resolution?


... after de-mosaicking?


Thanks.

 

[Graeme Nattress]

No I can't comment on Mysterium X.

If you make pixels and keep the same pixel architecture, yes, you may get a noisier pixel, but.... If you change (improve) the architecture as well....

Karapetov, I'd suggest they're good approximates until we can actually measure the camera. Chroma would be 1.5k though, luma nearer 2.4k perhaps?

Graeme

 

[Lauri Kettunen]

Graeme, in this thread you've given very useful information. Could you also share your thoughts about the IR-issue when heavy ND-filtering is employed?

 

[Graeme Nattress]

Lauri, I don't know enough about that to make the kind of educated posts I do about image processing.

Graeme

 

[Pawel Achtel]

Thanks Graeme for great explanations.

It is interesting what Sony claims to be advantages of their 3xCCD (in F23)over competitive CMOS designs:

1. More Stable Than CMOS Sensors
2. Repeatable Linear Response
3. Dependable Dynamic Range
4. No Bayer Filtering Required
5. Virtually No Fixed Pattern Noise
6. Noise Level Constant Through Full Dynamic Range

 

[Graeme Nattress]

2) I've measured Mysterium to be "very" linear, each and every time I do so.
3) is just another way of saying 2)
4) just means they're not using their photosite budget efficiently, and can't use standard cinema glass
5) not seen any here since it wasn't corrected in the first uses of Boris and Natasha for Crossing the Line
6) That's just sensors for you as read noise is constant no matter what level you read out
7) no vertical streaking - can that be said of the F23? That's the usual CCD defect
8) prism artifacts?
9) aliasing artifacts from inadequate OLPF? I've not pointed one at a zone plate myself, but maybe someone has and can report?

Graeme

 

[Pawel Achtel]

7) no vertical streaking - can that be said of the F23? That's the usual CCD

FIT CCDs do not suffer from vertical smear like IT CCDs do. 700A, F900, 750, 790 and f23 all use FIT CCDs. Some cheaper HDCAMs used IT sensors.

 

[Radoslav Karapetkov]

No I can't comment on Mysterium X.

If you make pixels and keep the same pixel architecture, yes, you may get a noisier pixel, but.... If you change (improve) the architecture as well....

Karapetov, I'd suggest they're good approximates until we can actually measure the camera. Chroma would be 1.5k though, luma nearer 2.4k perhaps?

Graeme

Thanks.

This thread is really important, maybe some of the mods should make it a sticky?

 

[Craig W. Bickerstaff]

4) just means they're not using their photosite budget efficiently, and can't use standard cinema glass

I would argue that for what they are trying to achieve with their sensors they are using it very efficiently. I would also argue that cinema glass is a very different issue and has little to do with which is the better system and is more a matter of taste.

 

[conrad gaunt]

The Red One sensor DOES do 4K from 1 to 60 fps.

You just can't record it. (at least, not yet)

Select "Open Gate" for your viewfinder, select 4K @ 30FPS, and what you're watching in your viewfinder is actually 4K at 60FPS. When in "open gate" mode, the sensor runs at twice the speed of the project's frame rate (meaning, 4K @ 24fps, the sensor is actually running at 48fps.)

Someday it may be able to record it, but right now you can see that the hardware does run at that speed. Don't know if that's enough to cover what the NAB PDF is claiming (and, hey, Build 16 isn't out yet -- maybe Build 16 includes 4K @ 60FPS?!)

I think Graeme actually said the R1 doesn`t have the processing power or bandwidth for 60fps@4k, which would require more than a sensor upgrade to put right. Maybe software optimizations could get it there, but reading a sensor and displaying it is different than reading a sensor, running it through a wavelet compression algorithm and outputting to a storage device. I`m not so hung up on the issue now, coming to terms with the idea. Yes, Red has spoilt me rotten. I hope R1 will be internally upgradeable in the future, making obselete, obselete etc

 

[Michael Hastings]

SLR glass is also possible and is a huge advantage.

SLR glass is also possible and is a huge advantage.

I would argue that for what they are trying to achieve with their sensors they are using it very efficiently. I would also argue that cinema glass is a very different issue and has little to do with which is the better system and is more a matter of taste.

But it is a matter of economics. Single Bayer sensor allows us to use SLR glass for just a few hundred dollars for excellent manual lenses to maybe $1700 for a new zoom with outstanding optical quality comparable to $25K or more for B4 glass required for 3 chip prism optical system.

Birger mount provides extremely smooth focus as well as use of most modern lenses.

While modern television zooms are optical marvels, I highly doubt that Canon would say that they provide sharper images than their own L series SLR lenses. Yet due to economies of scale the SLR lenses are a tenth of the cost.

 

[Dj Joofa]

With a 3 CCD approach you can apply RGB gain to each individual CCD which is something you cannot do with a single sensor. So not only are the 3 x mono sensors themselves more sensitive but you can colour balance to a certain degree within the sensor hardware (tungsten/daylight etc,.).

CMOS sensors have hardware on each pixel, does this hardware, in Reds case, allow for individual gain being applied to a pixel? Or do you feel that this isn't a huge issue for single sensors?

Typical (Bayer-based) CMOS sensors do provide mechanism for individual R,G and B gain, even though the photosites are on the same chip. You don't have to have 3 CMOS/CCD design to have individual R,G and B gain. This per R,G and B gain is a mechanism of providing analog gain, which is very useful in many cases. It appears that Red does not use the analog gain (though I am sure they have mechanism on their chip to provide it as any typical CMOS/Bayer sensor out there does that), and so some images acquired under certain settings may exhibit posterization because of a digital gain later in software.

 

[paulcurtis]

Typical (Bayer-based) CMOS sensors do provide mechanism for individual R,G and B gain, even though the photosites are on the same chip. You don't have to have 3 CMOS/CCD design to have individual R,G and B gain. This per R,G and B gain is a mechanism of providing analog gain, which is very useful in many cases. It appears that Red does not use the analog gain (though I am sure they have mechanism on their chip to provide it as any typical CMOS/Bayer sensor out there does that), and so some images acquired under certain settings may exhibit posterization because of a digital gain later in software.

Really? That's very interesting. Certainly CCD based single bayered sensors do not allow for separate gain (no surprise there though). I've hunted around for some cmos specs but not found anything that suggests separate RGB gain, only an overall one -- can you point me in a direction where i can find out more about this?

What makes you say that Red don't use it? I would have thought it would be vital under some lighting situations unless the cmos type of gain is not really the same as CCD analogue gain (and that doing it digitally afterwards would produce the same results)

I've certainly had issues with IT CCDs and blooming. i have yet to find any FIT based ones to play with though

cheers
paul

 

[Dj Joofa]

Really? That's very interesting. Certainly CCD based single bayered sensors do not allow for separate gain (no surprise there though). I've hunted around for some cmos specs but not found anything that suggests separate RGB gain, only an overall one -- can you point me in a direction where i can find out more about this?

Yes, many sensors from certain manufacturers, say Micron, provide analog gains (before ADC) for R, G, and B. Please note that the 3 gains are not per pixel, but per channel, i.e., individual R, G and B gains. But that is enough for a crude white balancing. As more typical white balancing involves a simple matrix multiplication. The difference in matrix multiplication will be that in addition to individual R,G,B gains, you can subtract/add contributions from other channels.

What makes you say that Red don't use it? I would have thought it would be vital under some lighting situations unless the cmos type of gain is not really the same as CCD analogue gain (and that doing it digitally afterwards would produce the same results)

Analog gain is analog gain, by definition provided before digitizing, i.e., before ADC. So the concept should stay the same among CCD and CMOS sensors.

I think it has been publicly stated on this forum many times that Red only records ISO/ASA setting in metadata instead of boosting signal using analog gain, and in fact does the gain later in software digitally.

 

[Randall Nott]

The main problem seems to be the blue filter passing some red light between 670nm and 700+nm. On film the blue layer does not record red light, but Bayer cameras may make deep red colors show as pink tinted or magenta.

Not meaning to hijack this thread, but the CMOS on my Sony A1 does this on an annoyingly regular basis when white balanced for indoors; to the point where I've stored a color correction preset for it in post. Suggestions for avoiding it at the aquisition stage (until I can afford my RED)?

Thanks Graeme, for reiterating vital info.