[BrianR]

Small photosites = less signal
less signal = more amplification
more amplification = more amplified noise
more amplified noise = poor signal to noise ratio
poor signal to noise ratio = more obvious noise in the image.

You left out:

Smaller photosites = more photosites sampled per unit area of final image

which works in favour of the higher mp sensor when viewing the images at the same size, such as 8x12 prints or both images scaled to fit your monitor (not viewed at 100%).

[normhead]

I disagree


Photosite size is a big factor.


The size of the photosites determines the light gathering capacity of the photosites. If the photosites are tiny and the sensor is therefore a high resolution sensor, which is what that's saying put slightly differently, the photons gathered by each photosite is therefore less, and the signal from each photosite, is therefore reduced and is lower.


The signal must then be amplified more to provide a coherent image because its weaker, the signal to noise ratio then changes because the greater magnification of the poorer signal amplifies the background noise as well, and more amplified noise is therefore generated. therefore the image carries more noise.


This surely is already well established.


Small photosites = less signal
less signal = more amplification
more amplification = more amplified noise
more amplified noise = poor signal to noise ratio
poor signal to noise ratio = more obvious noise in the image.


This is one reason why Full Frame sensors tend to have lower noise characteristics than smaller sensors, the photosites are bigger, they gather more photons, the signal is stronger and needs less amplification, the background noise doesn't get amplified as much and the noise is lower.


If you have a photosite twice the size it gather 4 times the photons and the background noise stays much the same, the signal is 4 times stronger and the signal to noise ratio is 4 times better, you then have much less noise in the image, due entirely to changing the photosite size and nothing else.


With real world losses, the truth is the difference isn't so pronounced as the raw figures would suggest but none the less noise is better controlled with larger photosites.

Actually this is not established at all... because sensors are a long way from optimizing the area each pixel takes on the sensor. With Sony's backlit sensor, and another sensor I can't remember the name of that puts all the circuitry on the back of the sensor leaving the whole front to collect photons... there is a loose correlation between pixel size and the number of photons collected, but it's far from written in stone.

A traditional, front-illuminated digital camera is constructed in a fashion similar to the human eye, with a lens at the front, wiring in the middle, and photodetectors at the back. This traditional orientation of the sensor places the active matrix of the digital camera image sensor -- a matrix of individual picture elements -- on its front surface and simplifies manufacturing. The matrix and its wiring, however, reflect some of the light, and thus the photocathode layer can only receive the remainder of the incoming light; the reflection reduces the signal that is available to be captured.[1]

A back-illuminated sensor contains the same elements, but orientates the wiring behind the photocathode layer by flipping the silicon wafer during manufacturing and then thinning its reverse side so that light can strike the photocathode layer without passing through the wiring layer.[6] This change can improve the chance of an input photon being captured from about 60% to over 90%

A back lit 24 Mp APS-c sensor could theoretically surpass a 24 MP FF sensor in photons collected. To make any kind of judgement about what sensor is collecting the most photons, you need to understand what technology is used in both sensors. Or as I usually say, as a general principal yes, you're right, looking at individual cases, you may be wrong.

In January 2012, Sony developed the back-side illuminated sensor further with Stacked CMOS, where the supporting circuitry is moved below the active pixel section, giving another 30% improvement to light capturing capability.[10] This was commercialized by Sony in August 2012 as Exmor RS with resolutions of 13 and 8 effective megapixels.[11] Products are expected in 2013.

I'm wondering if this is the type of sensor that's giving the 645z, Sony A7s and Nikon D4s their increased low light capacity, and if so, when it will make it to APS-c.

[robthebloke]

Just to re-enforce your argument, using pixel size or pitch, you have the following:

• Pentax K100D - 7.87 micron
• Pentax K10D - 6.07 micron
• Pentax 645D - 5.9 microns
• Pentax 645Z - 5.3 microns
• Nikon D800/e - 4.88 microns
• Pentax K5/II/IIs/30/50/500/-01 - 4.75 microns
• Pentax K3 - 3.9 microns
• Pentax Q - 1.5 microns

Looking at the 645Z, D800 and the K5 family. Their pixel size are all within 8% of each other

The pixels on the 645z are 25% larger than those on the k -5, not 8%. Microns are a linear measurement, whereas you should be using a squared measurement to compare them. (By comparison, pixel sizes on the 645z are 85% larger than the k-3)

[Rondec]

I think it is pretty hard to compare sensors. The biggest thing that I think larger pixels sizes gives (assuming similar tech) is bigger wells and therefore better dynamic range. A 645Z should be about two stops better in noise, just by nature of being a bigger sensor. That's probably what it is. Shrink that same sensor down to APS-C and you get K5 performance, roughly.

[Kayaker-J]

The pixels on the 645z are 25% larger than those on the k -5, not 8%. Microns are a linear measurement, whereas you should be using a squared measurement to compare them. (By comparison, pixel sizes on the 645z are 85% larger than the k-3)

Regarding common assumptions, the pixel site size on my Olympus m4/3 E-PL5 is nearly the same as on my K3. At the most commonly seen magnification ratios -- modest ones -- and at normal viewing distances, it might be more clear why Olympus has done so well in apple to apple test comparisons (of somewhat limited scope with regard to subjective factors, of course)... and in the consequent published ratings -- see 'Popular Photography' tests & ratings, for example. However, the m4/3 camera does seem to render what noise it does generate in what has been called more than once, a pleasing, "film-like" manner. This does not come from cranking up the usual noise reduction algorythms (Panasonic?), as high resolution and sharp acuity are retained. Multiple variables are involved in how all this turns out from a subjective viewpoint (objectively speaking); it's complex, making some generalizations commonly seen rather suspect in their value and utility.

[RonHendriks1966]

If the Sensor in the 645z is capable of 51 MP and exceptional high ISO performance, why isn't there a 24 Mp APS-c version of the sensor?
It would seem that such a sensor would knock the sox of the D4s or A7s with equivalent low light performance, while having 50% more resolution, and give us birders the best of both worlds. APS-c reach and FF low light performance.

Come on Sony, cut those wafers into smaller sizes....

Okay, I see a ball and need to kick it......


That is the talk about a custom sensor and from that same wafer you can make a nice 20-21 megapixel aps-h sensor. Sticking to aps-c would only give 13 megapixel. Some of the advantages of the big sensor are lost when going down to a smaller sensor.

We still haven't seen DxO test for the sensor. So don't make it more magical then it is.

[ElJamoquio]

This surely is already well established.

It's an internet myth. It's true for pixel-level noise but irrelevant in the era of 10+ MP pictures. You can't see an individual pixel.

This is one reason why Full Frame sensors tend to have lower noise characteristics than smaller sensors, the photosites are bigger, they gather more photons, the signal is stronger and needs less amplification, the background noise doesn't get amplified as much and the noise is lower.

The reason that full frame pictures have less noise (while holding ISO constant) is that the overall area of light-gathering goes up by a factor of 2.3. How many buckets ('photosites') you divide that area into is irrelevant.

---------- Post added 08-03-14 at 02:35 PM ----------

With Sony's backlit sensor, and another sensor I can't remember the name of that puts all the circuitry on the back of the sensor leaving the whole front to collect photons... there is a loose correlation between pixel size and the number of photons collected, but it's far from written in stone.

The lenses on the sensors make back-lighting less important than it otherwise might be, FWIW. Those lenses curve the photon around the wiring, or at least attempt to.

One of the areas that still could yield a 3x improvement is something like the Foveon, where in theory at least, no photon is wasted.

[Nicolas06]

As far as im concerned this is absolutely on the money and well said.


But there are I think 3 factors affecting noise in sensors.


1 is the photosite size

2 is the basic chip construction CMOS or CCD (or indeed foveon) which affects the signal to noise ratio and quality of the image.
3 is the age of the chip. - Old chips were much noisier. look at the 2 megapixel chips with large photosites they were much noisier, iso 800 on those old 2001 sensors was often noisier than iso 64,000 is on some sensors today.

From what we have seen until today from actual devices:

1 is the sensor size... By far the biggest factor here

As you can see on this image there there is rougly 7EV difference on light gathering between smallest and biggest sensor... And that's because we don't yet have large sensor like 8x10...



2 Is the sensor technology.

- Do you use CMOS or CCD, how technology evolve arround it, how much the surface is used for circuitry or light gathering... How much is lost in colors filter and so on.

We got mostly 2EV from this front in the 5 last years... Most of it is using CMOS instead of CCD. I have seen 3EV difference with ist* and latest K3 and this is 8 years evolution on sensors.

3 Software

Big progress has been made on noise removal software in particular like lightroom and DxO. You gain easily 1-2EV compared as without noise removal.

4 Photosite size... Depending on the design, the more photosites you have, the less part of the sensor surface is used for light gathering (mostly dependant of the technology). This is not really visible in practice on latest models.


Why I put photosite size last?

Well the A7s bet on bigger photosite size (3 time bigger surface than D800) but is only gaining 1/3 of stop on DxO while the size factor if linear could mean gaining 1.5stop. More: they clain they didn't just use bigger photosites but also an improved design. My understanding is that only part of the improvement comes from the bigger photosites... maybe half... 1/6 of an EV... That's not really noticable or impressive to be honest.

You can add more examples: D600 & D800 have similar high iso performance but not the same photosite size. K5 has roughly same photosite site than D800 and same technology inside... But D800 a little more than 2.3 time the high iso performance in DxO ... And the sensor that is... 2.3 time bigger.

Photosite size has marginal impact on high iso performance.

Why I put sensor size first?

Because well software is available to everybody... You just need to wait for it and use when available. This doesn't really count. But all the progress made on sensor efficiancy in maybe 10 year cover more or less the performance difference between m4/3 and FF... If you cover the whole spectrum of sensor size, you may need to wait 10 year more until your small sensor beat the oldest FF. For practicall purpose and with sensor technology not making big advence at the moment... if you want more high iso performance, you want a bigger sensor. And this is not the A7s you should look at if that important for you... But more 645Z. Again the sensor size.

[normhead]

I'm a little unclear on how a sensor cut from the same wafer would have less High ISO performance than a larger sensor. If you enlarge them both the same amount, they'll have the same noise...just the larger sensor image will be bigger. It could be argued that the larger sensor image will always have less noise when enlarged to the same size... but that assumes that there isn't a point at which noise becomes visible in the smaller format, which there clearly is. If the noise isn't visible in the smaller sensor image then functionally there will be no difference between the images in terms of noise or low light performance, at that size. It makes no difference if the noise from the larger sensor is better, if you can't see noise in either. So as per most things. putting sensor size first, needs some caveats. The first being that in images up to 8x10 on small format an 11x14 on APS-c it is quite possible there will be no difference at all, and that you'd have to go to much larger print sizes to notice any difference at all. And even then it would be debatable whether you'd see a difference form a "normal viewing " difference. As reported on the forum, if you get close to a D800 50' wide image you see more detail than on a k-5 image, stand back 4 feet, and the difference becomes less, or maybe even indistinguishable.

The 645z has much better ISO performance than a 645D just 4 years later. Apparenlty between 3 and 4 stops. The 645D had less Dynamic range and worse high ISO performance than a D800 with sensor 1.7 times larger. There is lot's of evidence out there, that cameras need to be evaluated case by case, not based on sensor size or any other arbitrary criteria. I only care about what actually is, not about what someone thinks "should" be true.

[Nicolas06]

I'm a little unclear on how a sensor cut from the same wafer would have less High ISO performance than a larger sensor. If you enlarge them both the same amount, they'll have the same noise...just the larger sensor image will be bigger. It could be argued that the larger sensor image will always have less noise when enlarged to the same size... but that assumes that there isn't a point at which noise becomes visible in the smaller format, which there clearly is. If the noise isn't visible in the smaller sensor image then functionally there will be no difference between the images in terms of noise or low light performance, at that size. It makes no difference if the noise from the larger sensor is better, if you can't see noise in either. So as per most things. putting sensor size first, needs some caveats. The first being that in images up to 8x10 on small format an 11x14 on APS-c it is quite possible there will be no difference at all, and that you'd have to go to much larger print sizes to notice any difference at all. And even then it would be debatable whether you'd see a difference form a "normal viewing " difference. As reported on the forum, if you get close to a D800 50' wide image you see more detail than on a k-5 image, stand back 4 feet, and the difference becomes less, or maybe even indistinguishable.

That why the reported noise metrics from DxO as an example look at the image quality you get by resizing it to 8MP the size needed for a 300dpi print of a standard A4 sheet of paper (21cmx29.7cm or rougly roughly 8.3"x11.5"). When comparing a 36MP D800 with a 51MP 645Z (not yet tested !) or a 12MP A7s... the comparison is always made at 8MP anyway(*).

(*) You can if you want look at the mesure without downscaling to 8MP, but as you explained theses measurement are not as informative in reality. What I care is that when I print my photo or when I look at it on my screen that the photo look good. I don't care much that if I look at a 100% crop the picture because this is not how a picture is seen in normal conditions.

Of course then it is likely that up to a point there is no visible difference when you bump the iso. Like in many situation on a K5 one doesn't see a difference between 100 and 800 isos except if he want to recover shadows or print at huge size and look carefully at the detail (maybe from a smaller viewing distance than one might think...). Still starting 3200iso on APSC for now noise is really visible just looking at the picture on screen with a mere 2MP... Like you can do the same on FF starting with 6400isos. There still room for improvement.

[Rondec]

You just need to assume the same tech when comparing cameras. Medium format tends to be updated much less frequently than full frame or APS-C, so the 645D was probably about the K10 level of tech, while the 645Z is more like the K5 level of tech.

I think the biggest reasons to move up in format size have to do with the ability to print larger, with more detail and less noise compared to smaller format. At web size, you may not be able to tell the difference at 36 inches, you probably can. At five feet on a side you definitely could.