BMCC4K? New Fairchild Sensor - Same DR of the Pocket and Cinema, but in 4.5k

[Brian@202020]

Is it a fact that BMD is using Fairchild and CMOSIS in their cameras ?, and where can I find the source of this information

It's common knowledge. You can find old threads on the Blackmagic forum confirming.

[Howie Roll]

So it's this one in the 4k and Ursa? http://www.cmosis.com/products/stand...ducts/cmv12000

Interesting how the dynamic range of Cmosis chip is 60db with extended dynamic range of 90db. I wonder if that's why there's such a drastic difference in DR between raw and ProRes on the 4k cams while the DR difference is negligible on the cinema cams.

I'd be surprised to see this chip on anything from BM this year, maybe NAB 2016. Looks great on paper.

[Tzedekh]

The Fairchild is also speedier. The CMV12000 can only do a maximum of 90fps with full image quality (URSA currently tops out at 80fps.) The Fairchild is listed at 240fps.

The CMV12000 maxes out at 180 fps at 12 bits and 300 fps at 10 bits. The CMV12000 has a global shutter. The LTN4625A has both rolling-shutter (RS) and global-shutter (GS) modes. It runs at 240 fps in RS mode and probably something like 120 fps in GS mode (the CIS2521's max GS frame rate is half that of RS). So, on paper, the CMOSIS chip wins on this spec. But the LTN4625A probably will win on dynamic range even in GS mode, where, if the CIS2521 is any indicator, it will lose about a stop but still be theoretically capable of about 13 stops.

[combatentropy]

As a global-shutter fan, I am keenly interested in how well this sensor works in global-shutter mode.

Sensors are so very close to my ideal these days:

1. Dynamic range: I think I need about 12 stops. The difference between something white and something black, under the same light, is 5 or 6 stops. The difference between sun and shade on a clear day is 3 to 5 stops. So if you take worst case of both, that's 6 + 5. This is from some rough tests with less than the best gear, but Roger N. Clark also came up with 11 stops in an extreme lighting situation. So 12 honest stops should get me through most situations.
   
   
2. ISO: Of course I would love ISO 409,000 or whatever the Sony A7S has. But I'm trying to be reasonable in my ISO requirements the way I am with DR. I really would like to shoot in moonlight, like they did in this video, which they say was shot at ISO 12800-32000. It doesn't have to be noise-free at those ISOs.
   
   
3. Frame rate: Honestly I think all I need is 60 fps, from what I've seen that Twixtor can do. I just need to shoot at a high enough frame rate to give Twixtor enough information to interpolate cleanly.

So we already have sensors with these specs, except when you add in global shutter. Even then, the CMOSIS comes very close. It's mainly in the ISO that it's not there yet.

[Steven Abrams]

The sensor in the BMPCC is described as having a dynamic range of >88 dB and this new sensor has 88 dB so it could prove to be a little less than the BMCC/BMPCC which is described as having 13 stops.

Dynamic range is obviously so much more than this number though. Otherwise the BMCC/BMPCC would have been 14.5 stops like Joshua says. Considering it's been 3 years or so since the BMCC, perhaps things have advanced so this sensor might actually make use of the whole range, and depending on optimizations with dark current/photon noise/read noise/ADC noise/etc etc possibly it could even do more! The BMCC/BMPCC sensors are also dual gain (assuming folklore is correct), so that could help too.

[LochnessDigital]

Total speculation and wishlist from me, but I hope they follow suit of JVC and put this sensor behind a m4/3 mount with variable crop modes.

If my math is correct, the Veydra Mini Primes from 16mm to 50mm would cover this sensor.

For a pixel size of 5.5 micrometers, that would give us an imaging size of 22.53 mm by 11.88 mm if we assume a 1:1 crop of the sensor in a 4096 x 2160 resolution. Via the Pythagorean theorem, this means we need an image circle of 25.47 mm to cover this area. At 3840 x 2160 we need an image circle of 24.23 mm.

The 25, 35, and 50 would cover it with flying colors (31, 31, and 33 mm image circles, respectively), and the 16mm would just barely squeeze by at 26mm. The 12mm would vignette as its image circle is 24mm. They're also working on an 85mm but have not reported the image circle dimensions of it yet.

A 2K center crop would be awesome for the C-mount and other 16mm lens dudes. A 3K crop would ensure no vignetting any m4/3 designed lenses that don't cover the full 4K. This would be in the ballpark of the Cinema cam's imaging area as well.

Wishful thinking, but if they did this, they could essentially merge the production camera and cinema camera together into one cam.

[Joshua Cadmium]

Dynamic range is obviously so much more than this number though. Otherwise the BMCC/BMPCC would have been 14.5 stops like Joshua says. Considering it's been 3 years or so since the BMCC, perhaps things have advanced so this sensor might actually make use of the whole range, and depending on optimizations with dark current/photon noise/read noise/ADC noise/etc etc possibly it could even do more! The BMCC/BMPCC sensors are also dual gain (assuming folklore is correct), so that could help too.

Well, the optimizations over the last few years have made it so that Fairchild can achieve 88dB with a smaller photosite size (as moving from 6.5um to 5.5um is about a 30% reduction in area.) It's difficult to have high dynamic range when the buckets that hold it aren't large. For instance, one reason why the Alexa has such high dynamic range is that the sensor uses a 8.25um photosite.

The BMCC/BMPCC does have about 14.5 stops, as measured. But "usable" stops is closer to 13. Usable stops is more subjective, based more on the individual and the quality of the chart/testing they are using.

The sensor in the BMPCC is described as having a dynamic range of >88 dB and this new sensor has 88 dB so it could prove to be a little less than the BMCC/BMPCC which is described as having 13 stops. Then subtract a full stop or so to enable global shutter and you're the same DR as the BMPC4K/URSA.

The BMCC has a dynamic range of 86dB, and the BMPCC has just around 88dB. It should be really close in rolling. If Blackmagic enabled global and it cost a stop, you'd be around 13.5 measured or 12 "usable" stops.

The BMPC4K/URSA actually has only 9.9 measured, or about 8.5 "usable" stops.

So it's this one in the 4k and Ursa? http://www.cmosis.com/products/stand...ducts/cmv12000

Interesting how the dynamic range of Cmosis chip is 60db with extended dynamic range of 90db. I wonder if that's why there's such a drastic difference in DR between raw and ProRes on the 4k cams while the DR difference is negligible on the cinema cams.

The 60dB comes from dividing the full well capacity by dark noise, which is the normal way of measuring sensor dynamic range.

Full well capacity is where the photosite is saturated by electrons (which is what the photons you are capturing are converted to) and the image is fully blown out. Dark noise is where electrons leak out of the small amount of electricity that is always present, and those electrons are naturally mistaken for photons.

After you divide the two, you would take the 20log10 of the result to arrive at the decibel number, which just describes the ratio between the two numbers in a logarithmic scale.

In this case, full well capacity is 13,500e and dark noise is 13e, so 13500/13=1038.46. Significant digits mean we have to take the result as 1000. 20log(1000) = 60db.

The 90dB can only work in an extended DR mode that combines frames at different exposure times, which leads to odd motion smearing. The Production/URSA do not use this mode.

[Steven Abrams]

The BMCC/BMPCC does have about 14.5 stops, as measured. But "usable" stops is closer to 13. Usable stops is more subjective, based more on the individual and the quality of the chart/testing they are using.

This was more my point. Wouldn't the work the camera manufacturer does (rather than the sensor manufacturer) also determine "useable" range, so perhaps the last 3 or so years could also possibly be used to increase the useable DR? BM rated those cameras as 13 stops, so that's what they must have considered useable at the time. That could also improve knowing that the BMCC was their FIRST camera and they're bound to have learnt a lot since then.

[combatentropy]

the LTN4625A probably will win on dynamic range even in GS mode, where, if the CIS2521 is any indicator, it will lose about a stop

Are you saying that you know that the CIS2521 in global-shutter mode loses only about a stop?

[combatentropy]

"Fairchild imaging is pleased to invite you to visit our booth at the Automate Trade Show, taking place at the McCormick Place Convention Center, Chicago, Illinois, United States, 23-26 March 2015. Come to our booth for a live presentation of the LTN4625A - our 12M pixel, 240 fps, 88dB high dynamic sCMOS Image Sensor! We will have people on hand to answer any questions you may have. Find us at booth# – 1196"

Anyone going?

[Abstrak]

If they implement this to the Ursa and give it a 800 base iso with a decent expanded iso range with little noise the Ursa will sell like hot cakes.

I'd upgrade my 4K to the URSA if they made the camera smaller and gave us 240 FPS 4K

[Tzedekh]

Are you saying that you know that the CIS2521 in global-shutter mode loses only about a stop?

According to pco's pco.edge 5.5 data sheet, the slow-scan rolling-shutter (RS) readout noise is 1.1 e- (median) or 1.5 e- (root mean square [rms]). The equation to derive dynamic range is

DR, dB = 20 x log(full-well capacity ÷ readout noise)

The stated full-well capacity is 30,000 e-. Using the median readout noise figure 1.1 e- yields an RS DR of 88.7 dB, and using the rms one gives 86.0 dB. So, it looks as if BAE/Fairchild are using the former to calculate DR, probably because it yields the higher figure, whereas they should probably be using the latter noise figure. According to one of Hamamatsu's "Reality Check" articles, "Read noise in CMOS cameras: only rms is meaningful" (here):

With CCDs there are never any issues regarding which model to use because the typical read noise for all pixels is very similar, thus rms and median are equivalent. With sCMOS, the structure of the sensor inherently has more pixel variation, and the extreme low noise of the sensor makes variation more statistically significant. So when it comes to evaluating camera performance, the truly meaningful spec is rms noise. The rms noise value provides insight into image quality as well as being the appropriate noise variable in quantitative calculations.

That means 86.0 dB is the more meaningful RS DR figure.

The median and rms readout noise specs for the fast-scan global-shutter (GS) mode are 2.2 e- and 2.5 e-, respectively, which yield DR figures of 82.7 dB and 81.6 dB. Allowing for rounding, the median-noise GS DR spec is one stop lower than the RS one, but the rms DR spec is less than one stop. Of course, these are theoretical figures -- I've honestly never seen any reports of real-world tests of the CIS2521's global-shutter dynamic range.

[combatentropy]

Thank you. I agree with you about the root mean square.

Either way, if this is the case, I don't know why Blackmagic didn't offer it in global shutter mode. Worst case, 81.6 dB, that's 13.6 stops.

[Tzedekh]

I don't know why Blackmagic didn't offer it in global shutter mode. Worst case, 81.6 dB, that's 13.6 stops.

Well, no, that's the best case. Maybe tests indicated that noise -- and therefore dynamic range -- were much worse in practice than these figures would predict. Or maybe implementing a global-shutter mode would have raised production costs beyond their targets or significantly delayed release.

[Steven Abrams]

http://www.fairchildimaging.com/cata...scmos/ltn4625a

"> 40,000 e- full well capacity" vs "> 30,000 e- full well capacity" for the Pocket - this should mean more DR than Pocket/BMCC !
"< 15 e-/pixel/sec at 20C" vs "< 30 e-/pixel/sec at 20C" for the Pocket cam - half the dark current should mean better noise performance which could mean EVEN more useable DR than Pocket/BMCC on top of the gain for more full well capacity !!

This is looking to be VERY exciting.