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

willie_901

willie_901

Veteran
Local time
6:38 PM
Joined
Dec 12, 2005
Messages
5,689
Recently Bill Claff published measurements of the X-Pro 2's read-noise level vs ISO. This method estimates the input read-noise level based on statistical analysis of unrendered raw data.

The results (link) were surprising. The X-Pro 2 data-stream design and performance is very different from the XTrans and XTrans II cameras'.

About a week later Thom Hogan discussed dual-gain sensor technology. Dual-gain engineering actually changes the capacitance of the sensor well above some ISO level (800 in the case of the X-Pro 2). This effectively boosts the signal level (DC voltage from each sensor site) with practically no increase in the read noise level. Above ISO 800 the DC signals are further amplified using conventional DC methods.

The X-Pro 2 has two base ISOs, 200 and 800.

Because the signal-to-noise ratio improves, the analog dynamic range improves. Claff also published data to that shows the X-Pro 2's dynamic arrange vs ISO is quite close to the theoretical maximum one can expect from an APS-C sensor at ISO 800 and above.

Here's the data.

To be complete, the recently announced SONY a6000 APS-C uses the same technology.

Still, the X-Pro 2 is more than just a larger MP version of the XTrans I and II cameras. The signal-to-noise ratio at ISO 800 and means shadow regions in low-light levels should render as well as possible for an APS-C platform.
 
That's very interesting. If you dig through the site, it shows that there is, broadly, no improvement in shadow range above iso80o until you reach the extended settings. Potentially worth using auto iso, limiting to 800 and manual control of aperture and shutter in low and difficult light situations. My impression of the matrix metering so far is very positive in terms of protecting highlights.

The only, very material, downside is that it makes the M9 look very weak and yet I still like the rangefinder focus and viewing approach for out and about. The M240 doesn't seem, to me, to offer enough in addition to make the jump worthwhile at this stage - unless paired with a Kolari A7r2

Mike
 
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.
 
CameraQuest said:
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.
Click to expand...

I would hope that four years later, using a different type of wiring, and holding the hand of one of the world's most sophisticated sensor manufacturer, Fuji would be able to beat CMOSIS. I've often thought that Leica's selection of the CMOSIS was either aimed at sidestepping a keiretsu or, perhaps more remotely, playing to eurocentrism.

But all of that said, the charts account for noise but not the signal. In general, larger photosites drown out noise by capturing more photons.

The dual-ISO technology is pretty cool stuff.

Dante
 
Dante_Stella said:
What's the signal part of the equation?
Click to expand...

The signal is the number of electron-hole pairs created in each sensor site by light amplitude absorption. This signal level is determined only by the exposure. The average signal level for the sensor is a function of shutter time and aperture. Nothing else affects how many of electron/hole pairs are created as ISO changes. That is the sensor sites' quantum efficiency is independent of ISO.

The above assumes there the number of electron-hole pairs does not exceed the full-well capacity of the sensor (over exposure). When this happens the signal becomes more complex because not all the sensor sites contains the same type of information content.

Next, the electrons are converted to electrical charge and that charge is converted to DC voltage. Typically the electron-hole pair creation is not mentioned and the story starts with the electrical charge stored at each sensor site.

When ISO is set above base ISO the analog DC voltage from each sensor site is amplified. But for dual-gain data streams DC amplification is not applied for the lowest ISO in the second gain stage (800 for the X-Pro 2). Otherwise, increasing the DC voltage level serves two purposes. With older analog-to-digital converter technologies, the ADC read-noise contribution is minimized. However the main puropose is to make the light meter's estimate come true. This is particularly important for those who use in-camera JPEGS.

While the number of electron hole pairs can not be increased after the shutter closes. The initial conversion to electrical charge depends on the site's capacitance. Changing the capacitance increases the stored electrical charge with the a lower level of read noise than amplifying a DC voltage.

When one obeys the light meter the exposure decreases as ISO increases. This is also the case for ISO 800 in a dual-gain camera. The difference is: compared to single stage gain, the read noise level is lower so the signal-to-noise ratio. increases.
 
Dante_Stella said:
I would hope that four years later, using a different type of wiring, and holding the hand of one of the world's most sophisticated sensor manufacturer, Fuji would be able to beat CMOSIS. I've often thought that Leica's selection of the CMOSIS was either aimed at sidestepping a keiretsu or, perhaps more remotely, playing to eurocentrism.

But all of that said, the charts account for noise but not the signal. In general, larger photosites drown out noise by capturing more photons.

The dual-ISO technology is pretty cool stuff.

Dante
Click to expand...


Agree there are several ways to measure. Possibly more useful is the second Cliff link, which shows 'photographic Dynamic Range's - a print normalised measure. The M240 is slightly ahead of the XP2 at low isosceles and slightly behind at high. The SL is slightly ahead throughout the range. The Sony A7R2 is materially better than all.

Downside is, I played with a previous gen Sony today and hated it.

Mike
 
Dante,

An increased photo-site area means more electron-hole pairs are created. This is good.

At the same time, the read-noise at ISO 800 is reduced by about a factor of 6. Assuming the quantum efficiencies are identical. The increase in area would have to be large enough for the single-gain sensor signal level to match the dual-gain sensor's SNR on a single pixel basis.

So, a pixel diameter increase of about 5 [(sq. root of 6) X 2] could generate enough signal at ISO 800 to offset the decrease in dual-gain sensor's read-noise advantage.
 
Dante_Stella said:
But all of that said, the charts account for noise but not the signal. In general, larger photosites drown out noise by capturing more photons.

Dante
Click to expand...

As Mike suggested above, if you look at the Bill Claff's dynamic range vs ISO data, you can get a idea for how the SNRs compare. DR is directly dependent on the total analog SNR.

The DR also depends on then total sensor area means 24 X 36 mm sensors have an inherent signal level advantage over APS-C sensors.
 
sojournerphoto said:
Agree there are several ways to measure. Possibly more useful is the second Cliff link, which shows 'photographic Dynamic Range's - a print normalised measure. The M240 is slightly ahead of the XP2 at low isosceles and slightly behind at high. The SL is slightly ahead throughout the range. The Sony A7R2 is materially better than all.

Downside is, I played with a previous gen Sony today and hated it.

Mike
Click to expand...

Actually, I must have been looking at the wrong link before. Looking at the PDR chart, the result is actually pretty close (and the difference either way, M240 or X-Pro2, is only 1/6 to 1/3 stop, which is within the margin of error for lens aperture markings).

The M240 does look like an underachiever in that it only performs up to the theoretical maximum for a DX camera - it, and almost all FX cameras, seem to be well below what FX should be able to do. This is good news for FX camera makers, who will at least be able to crank out newer models with newer sensors with more DR (the A7RII maybe not so much, since past ISO 32,000 it is already there). For DX, however, there's not really anywhere to go after the Sony/Fuji sensor. So maybe there is credence to rumors of a MF Fuji digital.

But when you look at it technologically, getting a DX sensor to perform like an FX one (even an underachiever) is like the exercise of putting a turbo on a 1.6L engine to get to the same horsepower as a small-block V8. It takes a lot of technology to overcome physics.

I was gratified to see at least that the M240 has the same DR as a D700 (which is great), just twice as many pixels.

Dante
 
Interesting article.

Dante_Stella said:
But when you look at it technologically, getting a DX sensor to perform like an FX one (even an underachiever) is like the exercise of putting a turbo on a 1.6L engine to get to the same horsepower as a small-block V8. It takes a lot of technology to overcome physics.
Click to expand...

Correct me if I'm wrong, but all technology made equal an FF camera should give one stop better SNR than an APSC sensor with the same number of pixels. That difference can be overcome with either fewer pixels, a faster lens, or a stop lower ISO/dynamic range performance. Not to say that FF doesn't have it's other benefits, but if we're talking pure IQ performance there's not a lot in it.
 
nongfuspring said:
Interesting article.



Correct me if I'm wrong, but all technology made equal an FF camera should give one stop better SNR than an APSC sensor with the same number of pixels. That difference can be overcome with either fewer pixels, a faster lens, or a stop lower ISO/dynamic range performance. Not to say that FF doesn't have it's other benefits, but if we're talking pure IQ performance there's not a lot in it.
Click to expand...

The key to the comparison is normalising the result for a given print size. Then you get about 1 stop more DR from an 'fx' as compared to an APS-C sensor, regardless of the number of pixels, but assuming equal technology in each case.

The XP2 is very close to the theoretical ideal dx performance already at iso800 up. It's no slouch below either. The M240 and SL are both pretty close to the XP2, rather than close to the 'perfect fx' performance line. The A7R2 is close to the ideal fx performance at high iso.
 
Trius said:
From a purely business perspective, why would Fuji develop an FX system?
Click to expand...

I can't think of a single reason.

Fujifilm might decide to skip 24 X 36 mm sensors and produce a consuer-priced medium format (whatever that means right now) system. This would require a significant investment in new lens development. After all, lens purchases leverage the investment in new body development.

Given the current state of the new camera market I will speculate they are simply going to offer the new XTrans III system in versions of their current line and continue to develop the Fujinon lens lineup.
 
Just tried to compare the dynamic range curves of XP2 and XT1.

http://www.photonstophotos.net/Charts/PDR.htm#Fujifilm X-Pro2,Fujifilm X-T1,Ideal DX

The curves are interesting. From base iso of the Xt1 (200) to Iso 640 the performance is the same. So the landscape photographer will get the same known and good dynamic range. That means Fuji managed to squeeze in more MP and achieve the same low iso DR.
At Iso800 the curves split apart and beginning with iso 1600 it's completely different. At Iso6400 the XT1 is even 1 Stop better than the XP2.

Does the split of the signal at Iso800 mean, they gave up the idea of a complete isoless sensor?
 
You must log in or register to reply here.
Back
Top Bottom