S35 Sensor with Higher Resolution

[roryhinds]

Is it possible to develop a higher resolution sensor and keep the physical size down to within S35?

Having a 6k S35 sensor would mean fantastic 4k images in a 4k postproduction workflow.

Is this possible with the current RED sensor design or do you need a new design and different technology?

[Andrew McCarrick]

Is it possible to develop a higher resolution sensor and keep the physical size down to within S35?

Having a 6k S35 sensor would mean fantastic 4k images in a 4k postproduction workflow.

Is this possible with the current RED sensor design or do you need a new design and different technology?

Could be done, but you're looking at smaller pixels, meaning less DR.

An analogy I came up with was a waffle with syrup... the smaller you make the pits on the waffle, the less syrup that can fit in each one, but you can put more pits on the total waffle... same idea with a pixel on a sensor. Smaller the pixel = less light = less DR = more pixels

[Clint Johnson]

There is also the problem of getting glass to resolve detail at that small photosite size. A 5K image is getting above the 100 line pairs per millimetre that even the best lenses can focus detail to. You should expect to get considerably more resolution with 6K FF35 than 6K S35 could give you.

For a full frame 35mm sensor, 9K is getting to the point where it is out resolving all but the very best lenses.

[Daniel Browning]

Could be done, but you're looking at smaller pixels, meaning less DR.


Actually, the opposite is more often true. Smaller pixels can always be resampled to the same size as larger pixels, which reduces their random noise by the square root of the quotient of the two areas. In general, smaller pixels tend to have higher QE/sq. micron and lower read noise per square micron, so they usually end up being slightly superior in the end. And that's not even taking into the account the effect of OLPF.

That's why, for example, the 1Ds Mark III has more dynamic range than the original 5D, despite having pixels that are much smaller.


There is also the problem of getting glass to resolve detail at that small photosite size. A 5K image is getting above the 100 line pairs per millimetre that even the best lenses can focus detail to.


Some lenses can do far better than that. For example, check out this $450 lens on 1.175 micron pixels (20K, 242MP APS-C / Super35 sensor) (http://forums.dpreview.com/forums/read.asp?forum=1019&message=29826265). The pixels were simulated by using 4X TC. The 100% crop clearly has usable detail.

http://i.pbase.com/t1/28/329428/4/105143389.eSU5fJtV.jpg (http://forums.dpreview.com/forums/read.asp?forum=1019&message=29826265)

[Clint Johnson]

Actually, the opposite is more often true. Smaller pixels can always be resampled to the same size as larger pixels, which reduces their random noise by the square root of the quotient of the two areas. In general, smaller pixels tend to have higher QE/sq. micron and lower read noise per square micron, so they usually end up being slightly superior in the end. And that's not even taking into the account the effect of OLPF.

That's why, for example, the 1Ds Mark III has more dynamic range than the original 5D, despite having pixels that are much smaller.



Some lenses can do far better than that. For example, check out this $450 lens on 1.175 micron pixels (20K, 242MP APS-C / Super35 sensor) (http://forums.dpreview.com/forums/read.asp?forum=1019&message=29826265). The pixels were simulated by using 4X TC. The 100% crop clearly has usable detail.

http://i.pbase.com/t1/28/329428/4/105143389.eSU5fJtV.jpg (http://forums.dpreview.com/forums/read.asp?forum=1019&message=29826265)

That methodology was very flawed. To take a lens that probably can't resolve much over 100 lp/mm by itself, stack three teleconverters behind it, put it on a midlevel camera with 4.7 micron photosites and take a picture of some wood slivers - then extrapolate that this "proves" that some lenses can show detail down to 1.175 micron photosites?

A teleconverter usually adds at least seven glass elements along with their attendant 14 surface transition phases and each one will degrade the light hitting it, getting worse and worse as it travels through this torture test of glass.

I don't think that a meticulously tweeked Zeiss Master Prime that managed to resolve over 200 lp/mm by itself will still resolve much over 100 lp/mm if it were stacked in front of a single teleconverter that was custom built and matched to that very lens.

John Sheehy seems to believe that his test proves that this $450 lens can somehow resolve over 400 line pairs per millimeter. If he had such a lens, it would be a freak accident of manufacturing that any lens maker would pay handsomely to study.

With the glass we actually can use, I'm pretty sure that 5 micron is getting down to the lower limit for photosites. I also feel confident that it is only a matter of time (and this could be in the lab right now) before someone comes up with a lens design paradigm that throws all this out the door and the hard physics imposed by the wavelength of light becomes the limiting factor.

[Daniel Browning]

That methodology was very flawed.

Would you explain why it is flawed?

I must have misunderstood your post, because it sounds like what you're saying is that his results are impossible. Do you think they were faked?

[Clint Johnson]

To test if a lens can resolve down to 1.175 micron photosites, it is best to use a sensor that has 1.175 micron photosites and a chart that has lines on it that progress to closer together than you are expecting the lens resolve to (this would be a factor of the actual line size and distance from the camera - finer lines closer can be equalled by coarser lines further away).

He used an inappropriate sensor, stacked three image distorting teleconverters on the lens then took pictures of an inappropriate target. You can't extrapolate this to a useful conclusion when none of the parameters relate to what was supposed to be tested.

The image he took showed wood fibres, which could be of any size, that faded in and out of visibility and had distortion, aberration and jpeg compression artifacts extending over fourteen or fifteen pixels. What would that have looked like if there were seven line pairs trying to resolve in the same area that one of those slivers distorted over?

He claimed this showed the lens to be able to resolve over 400 line pairs per millimetre when from what I see, I would expect it to extinguish far below that and show a uniform grey field with anything approaching even 200 line pairs per millimetre.