Are bigger pixels less noisy?

Started 5 months ago | Discussions thread
57even Forum Pro • Posts: 15,606
Re: Are bigger pixels less noisy?

bobn2 wrote:

57even wrote:

bobn2 wrote:

57even wrote:

bobn2 wrote:

57even wrote:

I don't think that - at moderately high ISO - pixel size makes a significant difference. Or rather, the design of the pixel makes a bigger difference than the size. Read noise is quite unpredictable and doesn't correlate all that well with sensor size.

I think the problem with the approach that you are arguing against is the desire to find absolute rules such as 'bigger pixels means less noise'.

Where was I doing that?

The actuality of designing sensors is one of trading off many different design parameters, and it is not a given that one design parameter on its own only gives benefits. Let's take pixel size, for instance. A big pixel might improve the proportion of incident light captured due to a higher fill factor. But then it will need a higher saturation capacity if it is to operate at the same exposure. That means a lower conversion gain which will increase read noise. Then the potential gradients in a large pixel will be shallower, so it might be more difficult to transfer the charge to the read gate, so the conversion efficiency might be lower than a small pixel. The only way to make generalised statements is to look for trends in performance data, and those don't show significantly better noise performance for bigger pixels.

Where did I say it did?

Sorry, a misunderstanding here. I wasn't disputing anything you said, I was backing up your position. I thought I'd made that clear when I said 'the problem with the approach that you are arguing with',

That could be taken either way when you think about it...

Yes, I meant it in the sense that the problem was with what he was saying, not what you were saying.

rather than 'the problem with your argument'. Obviously not clear enough. Sorry about that.

No problem. What you said about conversion gain is also true, but set that against larger source followers having less 1/f noise, and DCG which nullifies the gain issue.

Not really. I know that's what Eric Fossum said, but he's used to working at the bleeding edge, where 1/f noise is tghe major remaining 'read noise'. For current commercial cameras it's not the dominant component of read noise.

Not at low ISO, for sure. Which rather makes all the fanfare about DR rather meaningless. With FF at least, read noise isn't really a consideration in terms of exposure flexibility unless it's anisotropic. High ISO is a different issue.

However, there are interesting differences in some cases between dedicated low res sports cameras and their high-res landscape siblings, where the former have superior DR at high ISO (sometimes significant) and worse DR at low ISO (also significant).

A7S vs A7R4

https://www.dxomark.com/Cameras/Compare/Side-by-side/Sony-A7S-versus-Sony-A7R-IV___949_1326

Yes, I suspect that when you're talking about sensors made at the same time by the same manufacturer, you're looking at pixels designed form a CAD library and the basic read circuitry is optimise somewhere in the middle of the range of pixel sizes it will be used. That means that the per pixel read noise is fairly consistent between pixels of different sizes, and thus the sensor with fewer pixels wins on the read noise front, but the sensor with more pixels wins on the DR front.

D3s vs D3x

https://www.dxomark.com/Cameras/Compare/Side-by-side/Nikon-D3X-versus-Nikon-D3s___485_628

However, the D3 was no better at high ISO than the D3x, so this seems to be more to do with electronics than pixel pitch.

I don't think that the D3s and D3x is of much interest, because their sensors were completely different top to bottom. Different process, different pixel architecture, different read chain, different ADC architecture. Between the D3 and D3s Nikon made some very radical changes. First they put in two element microlenses, which solved one of the problems of big pixels, making microlenses that are effective, and they manage to reduce the pixel read noise by a factor of three or so. They essentially got it to Canon standards, which Sony was quite a long way behind at the time.

Yes, it was impressive. But it rather makes the point that pixel size isn't a useful indication of anything very much. The technical design, and the relevant compromises made for the intended use base, is a far more significant factor.

The low ISO issue seems to relate to gain and downstream noise - the high ISO difference seems to imply lower upstream noise (as a proportion of signal when normalised).

Yes, I think that's about it. As the read chain gain is raised, the downstream noises become more and more insignificant and you're left with the electronic noise at the front end of the sensor.

Which can become a significant factor in SNR at the top ISO ranges.

Or some massaging of the data in camera.

BTW, spoke to a friend who used to work for Pentax. He reckons they stuck with SOS to match their light meters at the same ISO (which they apparently do).

Can't say why Olympus or Fuji do it, but it's notable that they both use a much higher IE than the equivalent ISO speed. (ISO 200 vs 100 for instance). This should also align pretty well with a light meter with a K value of 14 (eg a Pentax one).

If they didn't calibrate their meters to SOS, you wouldn't be able to use the camera with an external meter without a correction, which would be problematic. I think the move to REI doesn't come from metering changes. it comes from the possibility of using of tone curves which move the 18% point to some other lightness value, as you might do for creative modes, and if you change the JPEG defaults. Rather than make the declared ISO standard switch from SOS to REI depending on the precise camera settings, it's easier just to claim REI. I think most cameras operate pretty close to SOS in their default modes.

Well, that would make sense, or their images would look very different. But Fuji don't change ISO in their various film simulations or when using evaluative modes.

I think the other agenda is that they don't want to state their camera's 'true' ISO speed because it would seem rather slow (at the high end as well). I don't know of any manufacturers that report it any more.

I gather 2019 version has a couple of new standards, one based on CIE Lab L* = 50 rather than Y = 118. Now that would be interesting, and pertinent given that its the standard profile connection space. Not raw of course, but less restrictive than sRGB from an output POV.

That said, I doubt most people would pay any notice.

-- hide signature --

"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away." Antoine de Saint-Exupery

 57even's gear list:57even's gear list
Fujifilm X-Pro2
Post (hide subjects) Posted by
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
(unknown member)
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow