X-Pro 2 Uses On-Sensor, Dual-Gain Technology

[willie_901]

...
Does the split of the signal at Iso800 mean, they gave up the idea of a complete isoless sensor?

Interesting question.

In a sense they did not. An ISO invariant data stream means the read-noise level due to DC amplification adds very little, if any, noise to the signal. So digital multiplication of base-ISO, raw data has no obvious impact on IQ. This is particularly apparent in shadow regions.

In the dual-gain case at ISO 800 the data is not amplified by a conventional DC amplifier. Instead the capacitance of the photo site itself is changed which increases the electrical charge stored when the shutter is open. The signal level does not increase (ISO 800 always produces less signal than ISO 200 when you do what the meter tells you to do). But boosting the signal level at the photo site generates less read noise than conventional designs.

I guess one way to think about this is the camera has two, ISO-less ranges – one below 800 and one above 800.

It would be interesting so compare shadow regions in an ISO 200 image with a 3 stop push to a ISO 800 image with no push. I will speculate there has to be some advantage or the only purpose for the expense of a dual-gain system (a licensing fee is paid to Aptina) would be for in-camera JPEG users. We do know the DR of at 800 and above approaches the maximum limit for APS-C sensors. Since the DR is computed form statistical analysis of un-rendered raw data, it seems reasonable to assume raw file shadow regions will benefit.

When DR is important, nothing beats base ISO. With low-light scenes, it appears signal-to-noise ratio at ISO 800 will be hard to beat.

 

[adamjbonn]

So... And please forgive my vague understanding of this...

The previous 2 versions of the X-Trans sensor used a mixure of analogue and digital gain up to 1600, then purely digital for 1600-6400

Theoretically meaning there's no real difference in shooting at 1600 and pushing 2 stops in post, or selecting 6400 in camera

But the X-Trans 3 (sic) sensor is 100% digital gain, truly "ISOless" (again, sic)

Implying that one could theoretically shoot everything at 200, then push 6 stops to get the same image brightness as 12800

But in reality, the amount of digital gain in the X-Pro2 has two floors, 200 and 800

So an 800 ISO shot pushed 4 stops would be cleaner than a 200 shot pushed by 6 stops? (Which I appreciate is probably try IRL anyway as 6 stops is a big ask!)

Is this the reason why Fuji have given us +/- 5 stops on the EV dial?


Notes:

Yes there's more to playing the ISOless game than pushing global exposure in post
6 stops of push is not within the capabilities of most PP SW... Ergo use of the word theoretical

 

[willie_901]

But the X-Trans 3 (sic) sensor is 100% digital gain, truly "ISOless" (again, sic)

I'm not sure the X-Pro 2 is truly ISO invariant. The DR vs ISO data indicates pure digital signal amplification (multiplication actually) starts at very high ISOs.

Implying that one could theoretically shoot everything at 200, then push 6 stops to get the same image brightness as 12800

This is the case for global luminance. But it for for the XPro-2 the signal-to-noise ratio would be lower than using ISO 800 and increasing brightness in post-production. The dual-gain sensor technology actually increases the signal at ISO 800 and above because the full-well capacity increases. The FWC increases because each pixels capacitance increases so more electrical charge can be stored. The read noise remains essentially constant.

But in reality, the amount of digital gain in the X-Pro2 has two floors, 200 and 800

Yes. There are two base ISOs.

So an 800 ISO shot pushed 4 stops would be cleaner than a 200 shot pushed by 6 stops? (Which I appreciate is probably try IRL anyway as 6 stops is a big ask!)

It should be (especially for shadow regions) since the SNR is higher.

Is this the reason why Fuji have given us +/- 5 stops on the EV dial?

I will speculate the increase to +/- 5 stops has to to do with customer expectations rather than technology.

I relied extensively on exposure blending for commercial residential photography. With the D700 I found blending 5 exposures produced better results than 3. When I switched to a XT-1 for the same work I manually bracketed 5 exposures. Before too long I realized I only needed three exposures. My thinking is: since the analog dynamic range of the X-T1 is higher than the D700's, I achieved the same range of exposure with only three brackets.

 

[adamjbonn]

Thanks Willie,

I had a quick and dirty play (!) last night using only the SOOC output and the internal RAW convertor, and my ISO800 pushed 3 stops shot*, was tangibly more noisy than my SOOC ISO 6400 one. It (the 800+3 shot) did contain more details though...

I need to make a proper test, and process the file as I would intend rather than simply letting the camera add 3 stops!

*with equal aperture and SS values, and jpg settings (obviously!)

 

[willie_901]

The Fujifilm in-camera JPEG rendering engine automatically applies additional noise filtering at high ISOs. This can not be modified by the photographer.

This is not the case for rendered raw files.

 

[SaveKodak]

I look forward to a good test of the M240 sensor compared to the Xpro2.

At least one Fuji insider considers the Xpro2 sensor the better of the two.

This should be no surprise since the M240 sensor was a generation behind on the day of it's release. I love my 240 but man, I wish they would quit with the posturing and start buying Sony sensors in whatever spec Leica needs. Even the Q has high ISO banding. Noise I can handle, banding I can not. Imagine an M with the 12mp A7S sensor and a thin Leica filter stack. Now THAT would be a reportage machine. Ditto for one with even the old 36mp sensor.