Are bigger pixels less noisy?

Started 5 months ago | Discussions
Iliah Borg Forum Pro • Posts: 28,665
Re: Are bigger pixels less noisy?
3

JahnG wrote:

how about the combined read noise of 50M pixels compared to the combined read noise of 20M pixels.

The combined read noise can be marginally, and I mean marginally, higher than the noise of a single pixel. Decreasing the maximum charge opens a possibility to improve performance in terms of noise.

Mr. Claff kindly keeps the original Aptina white paper on dual gain, it might help explaining the relationships: https://www.photonstophotos.net/Aptina/DR-Pix_WhitePaper.pdf

-- hide signature --
spider-mario
spider-mario Contributing Member • Posts: 910
Re: Are bigger pixels less noisy?

JahnG wrote:

But how about the combined read noise of 50M pixels compared to the combined read noise of 20M pixels. One might thus think that the sensor having only 40% pixel count would have much less combined read noise? (Or shouldn't we sum read noise?).

One would typically sum it in quadrature (sqrt(RN₁² + RN₂² + …)). So if you combine k pixels, each with read noise RN, into one combined pixel, you would consider it to have read noise sqrt(k · RN²) = sqrt(k) · RN.

So, if you apply this to an α7R IV at base ISO to reduce it from 61 MP to 24 MP, you arrive at sqrt(61/24) · 2.9 ≈ 4.6 e⁻ of read noise for the 24 million “new” pixels, which compares favorably to the 6.4 e⁻ of an α7 III and its native 24 MP.

At high ISO, the situation reverses somewhat: at, say, ISO 3200, you would be looking at sqrt(61 / 24) · 1.1 ≈ 1.8 e⁻ for an α7R IV reduced to 24 MP, or ~1.2 e⁻ for an α7 III.

 spider-mario's gear list:spider-mario's gear list
Canon G1 X III Olympus TG-6 Olympus E-M1 II Olympus E-M1 III Canon EOS R6 +15 more
Enginel Contributing Member • Posts: 863
Re: Are recent sensors less noisy?
1

J A C S wrote:

To replicate my 50/1.2 on FF, you would need something like an 8.9/0.21 lens, which would be faster than the theoretical limit of f/0.5, not to mention the practical one f/1 or so.

f/0.5 is in the air. You just need to have rear element touching the sensor to have high-index (of 2.38) to be in contact with sensor, making f/0.21 possible.

 Enginel's gear list:Enginel's gear list
Pentax Q Sony Alpha NEX-5N Sony E 18-55mm F3.5-5.6 OSS Sony E 50mm F1.8 OSS
JahnG
OP JahnG Veteran Member • Posts: 3,339
Re: Are bigger pixels less noisy?

Iliah Borg wrote:

JahnG wrote:

how about the combined read noise of 50M pixels compared to the combined read noise of 20M pixels.

The combined read noise can be marginally, and I mean marginally, higher than the noise of a single pixel. Decreasing the maximum charge opens a possibility to improve performance in terms of noise.

Mr. Claff kindly keeps the original Aptina white paper on dual gain, it might help explaining the relationships: https://www.photonstophotos.net/Aptina/DR-Pix_WhitePaper.pdf

Thank you for your comments and for the Aptina white paper.

I'm not sure but I think that I have been told that (newer?) Sony sensors might utilize the Aptina aproach of using both HCG and LCG?  But even if that is the case, probably not the sensor of my 6 years old DSLR, the Pentax K-S1 APSC camera?  (BTW, if I didn't get it wrong i understood that Aptina finds it better to use larger pixels?)

 JahnG's gear list:JahnG's gear list
Canon PowerShot A650 IS Panasonic ZS100 Pentax K100D Super Pentax K-r Pentax K-S1
JahnG
OP JahnG Veteran Member • Posts: 3,339
Re: Are bigger pixels less noisy?

spider-mario wrote:

JahnG wrote:

But how about the combined read noise of 50M pixels compared to the combined read noise of 20M pixels. One might thus think that the sensor having only 40% pixel count would have much less combined read noise? (Or shouldn't we sum read noise?).

One would typically sum it in quadrature (sqrt(RN₁² + RN₂² + …)). So if you combine k pixels, each with read noise RN, into one combined pixel, you would consider it to have read noise sqrt(k · RN²) = sqrt(k) · RN.

So, if you apply this to an α7R IV at base ISO to reduce it from 61 MP to 24 MP, you arrive at sqrt(61/24) · 2.9 ≈ 4.6 e⁻ of read noise for the 24 million “new” pixels, which compares favorably to the 6.4 e⁻ of an α7 III and its native 24 MP.

At high ISO, the situation reverses somewhat: at, say, ISO 3200, you would be looking at sqrt(61 / 24) · 1.1 ≈ 1.8 e⁻ for an α7R IV reduced to 24 MP, or ~1.2 e⁻ for an α7 III.

Interesting, thank you

 JahnG's gear list:JahnG's gear list
Canon PowerShot A650 IS Panasonic ZS100 Pentax K100D Super Pentax K-r Pentax K-S1
J A C S
J A C S Forum Pro • Posts: 19,272
Re: Are recent sensors less noisy?
1

Enginel wrote:

J A C S wrote:

To replicate my 50/1.2 on FF, you would need something like an 8.9/0.21 lens, which would be faster than the theoretical limit of f/0.5, not to mention the practical one f/1 or so.

f/0.5 is in the air. You just need to have rear element touching the sensor to have high-index (of 2.38) to be in contact with sensor, making f/0.21 possible.

I am sure that the creatures on those UFOs have even better devices.

alanr0 Senior Member • Posts: 2,472
Pixel size comparison & Aptina DCG
3

JahnG wrote:

Iliah Borg wrote:

JahnG wrote:

how about the combined read noise of 50M pixels compared to the combined read noise of 20M pixels.

The combined read noise can be marginally, and I mean marginally, higher than the noise of a single pixel. Decreasing the maximum charge opens a possibility to improve performance in terms of noise.

Mr. Claff kindly keeps the original Aptina white paper on dual gain, it might help explaining the relationships: https://www.photonstophotos.net/Aptina/DR-Pix_WhitePaper.pdf

Thank you for your comments and for the Aptina white paper.

I'm not sure but I think that I have been told that (newer?) Sony sensors might utilize the Aptina aproach of using both HCG and LCG? But even if that is the case, probably not the sensor of my 6 years old DSLR, the Pentax K-S1 APSC camera? (BTW, if I didn't get it wrong i understood that Aptina finds it better to use larger pixels?)

My take from the Aptina paper is that dual conversion gain allows large pixels to be used without impairing low ISO dynamic range.

It may help to take a closer look at Bill Claff's results for the Sony cameras reference by spider-mario.

https://www.photonstophotos.net/Charts/RN_e.htm#Pentax%20K-S1_12,Sony%20ILCE-5000_14,Sony%20ILCE-7M3_14,Sony%20ILCE-7RM4_14,Sony%20ILCE-7SM3_14

Note that the vertical scale is base 2 logarithmic, so each division is a factor of 2.

I have added the results for your Pentax K-S1. Results for the 2014 vintage Sony A5000 are very similar, with around 3 electrons RMS read noise around ISO 1000. The curves diverge where noise-reduction kicks in at ISO 5000 for the Pentax (open symbols).

The more recent Sony cameras all show much less high ISO read noise, at around 1 e- rms. They also show a steep step in noise near the unity gain ISO, where the gain corresponds to 1 ADC unit for each electron captured.

The step is where the dual conversion gain switches in. At lower gains, voltage noise from the source follower transistor connected to the floating diffusion is pretty much the same, but conversion gain is reduced by the additional capacitance switched in (Aptina paper, figure 5). The input-referred noise increases by an amount proportional to the extra capacitance.

The increase is largest for the large pixel 12 Mp A7Siii (which needs a greater capacitance to avoid saturation at ISO 100), lowest for the small pixel 61 Mp A7Riv, with the 24 Mp A7iii intermediate.

Sony A7Siii (12 Mp, 8.4 µm pixels, FWC 187 ke-, 10.8 e- rms noise at ISO 200).

Sony A7iii (24 Mp, 5.96 µm pixels, FWC 96 ke-, 5.4 e- rms noise at ISO 200)

Sony A7Riv (61 Mp, 3.76 µm pixels, FWC 36 ke-, 2.9 e- rms noise at ISO 200).

The ISO 200 performance of the Pentax K-S1 and Sony A5000 fall between the A7iii and A7Riv results, but input-referred noise rises more steeply at lower ISO.

Pentax K-S1 (20 Mp APS-C, 4.28 µm pixels, FWC 28 ke-, 3.7 e- rms noise at ISO 200)

Full well capacity and other parameters derived from DxO results on Bill's site are here.

Down-sampling to similar pixel count

If you are shooting at low ISO, and down-sampling to 6-12 Mp, then the A7Riv should deliver better dynamic range and sharpness.

Assuming noise adds in quadrature, I estimate 6.5 e- rms input-referred noise at ISO 200 after down-sampling from 61 Mp to 12 Mp, compared with 10.8 e- from the A7Siii.

In practice, there may be a futher improvement, since the down-sampled image will need less sharpening to compensate for Bayer interpolation.

To summarise and simplify for these 5 specific cameras:

  • At low ISO (100 - 200) larger pixels are noisier in term of input-referred noise per pixel.
  • With dual conversion gain, there is less variation of high ISO noise with pixel size. Larger, fewer pixels, gives lower noise per unit area.
  • More recent cameras have lower noise than older cameras.

If you are shooting at high ISO and only need 12 Mp or less, then the large pixel Sony A7Siii wins.

If you need high resolution, then high pixel count is the way to go.

-- hide signature --

Alan Robinson

robert1955 Veteran Member • Posts: 6,003
Re: Are recent sensors less noisy?

Enginel wrote:

J A C S wrote:

To replicate my 50/1.2 on FF, you would need something like an 8.9/0.21 lens, which would be faster than the theoretical limit of f/0.5, not to mention the practical one f/1 or so.

f/0.5 is in the air.

I don't know what John Paul Young thinks about that

You just need to have rear element touching the sensor to have high-index (of 2.38) to be in contact with sensor, making f/0.21 possible.

Just touch?

 robert1955's gear list:robert1955's gear list
Fujifilm X-E3 Fujifilm 16-55mm F2.8R LM WR Fujifilm XF 70-300 F4-5.6 R LM OIS WR
Iliah Borg Forum Pro • Posts: 28,665
Re: Are bigger pixels less noisy?
1

JahnG wrote:

Iliah Borg wrote:

JahnG wrote:

how about the combined read noise of 50M pixels compared to the combined read noise of 20M pixels.

The combined read noise can be marginally, and I mean marginally, higher than the noise of a single pixel. Decreasing the maximum charge opens a possibility to improve performance in terms of noise.

Mr. Claff kindly keeps the original Aptina white paper on dual gain, it might help explaining the relationships: https://www.photonstophotos.net/Aptina/DR-Pix_WhitePaper.pdf

Thank you for your comments and for the Aptina white paper.

It's not just about Sony sensors, it's also about noise characteristics dependence on conversion gain and charge capacity.

-- hide signature --
JahnG
OP JahnG Veteran Member • Posts: 3,339
Re: Pixel size comparison & Aptina DCG

alanr0 wrote:

JahnG wrote:

Iliah Borg wrote:

JahnG wrote:

how about the combined read noise of 50M pixels compared to the combined read noise of 20M pixels.

The combined read noise can be marginally, and I mean marginally, higher than the noise of a single pixel. Decreasing the maximum charge opens a possibility to improve performance in terms of noise.

Mr. Claff kindly keeps the original Aptina white paper on dual gain, it might help explaining the relationships: https://www.photonstophotos.net/Aptina/DR-Pix_WhitePaper.pdf

Thank you for your comments and for the Aptina white paper.

I'm not sure but I think that I have been told that (newer?) Sony sensors might utilize the Aptina aproach of using both HCG and LCG? But even if that is the case, probably not the sensor of my 6 years old DSLR, the Pentax K-S1 APSC camera? (BTW, if I didn't get it wrong i understood that Aptina finds it better to use larger pixels?)

My take from the Aptina paper is that dual conversion gain allows large pixels to be used without impairing low ISO dynamic range.

It may help to take a closer look at Bill Claff's results for the Sony cameras reference by spider-mario.

https://www.photonstophotos.net/Charts/RN_e.htm#Pentax%20K-S1_12,Sony%20ILCE-5000_14,Sony%20ILCE-7M3_14,Sony%20ILCE-7RM4_14,Sony%20ILCE-7SM3_14

Note that the vertical scale is base 2 logarithmic, so each division is a factor of 2.

I have added the results for your Pentax K-S1. Results for the 2014 vintage Sony A5000 are very similar, with around 3 electrons RMS read noise around ISO 1000. The curves diverge where noise-reduction kicks in at ISO 5000 for the Pentax (open symbols).

The more recent Sony cameras all show much less high ISO read noise, at around 1 e- rms. They also show a steep step in noise near the unity gain ISO, where the gain corresponds to 1 ADC unit for each electron captured.

The step is where the dual conversion gain switches in. At lower gains, voltage noise from the source follower transistor connected to the floating diffusion is pretty much the same, but conversion gain is reduced by the additional capacitance switched in (Aptina paper, figure 5). The input-referred noise increases by an amount proportional to the extra capacitance.

The increase is largest for the large pixel 12 Mp A7Siii (which needs a greater capacitance to avoid saturation at ISO 100), lowest for the small pixel 61 Mp A7Riv, with the 24 Mp A7iii intermediate.

Sony A7Siii (12 Mp, 8.4 µm pixels, FWC 187 ke-, 10.8 e- rms noise at ISO 200).

Sony A7iii (24 Mp, 5.96 µm pixels, FWC 96 ke-, 5.4 e- rms noise at ISO 200)

Sony A7Riv (61 Mp, 3.76 µm pixels, FWC 36 ke-, 2.9 e- rms noise at ISO 200).

The ISO 200 performance of the Pentax K-S1 and Sony A5000 fall between the A7iii and A7Riv results, but input-referred noise rises more steeply at lower ISO.

Pentax K-S1 (20 Mp APS-C, 4.28 µm pixels, FWC 28 ke-, 3.7 e- rms noise at ISO 200)

Full well capacity and other parameters derived from DxO results on Bill's site are here.

Down-sampling to similar pixel count

If you are shooting at low ISO, and down-sampling to 6-12 Mp, then the A7Riv should deliver better dynamic range and sharpness.

Assuming noise adds in quadrature, I estimate 6.5 e- rms input-referred noise at ISO 200 after down-sampling from 61 Mp to 12 Mp, compared with 10.8 e- from the A7Siii.

In practice, there may be a futher improvement, since the down-sampled image will need less sharpening to compensate for Bayer interpolation.

To summarise and simplify for these 5 specific cameras:

  • At low ISO (100 - 200) larger pixels are noisier in term of input-referred noise per pixel.
  • With dual conversion gain, there is less variation of high ISO noise with pixel size. Larger, fewer pixels, gives lower noise per unit area.
  • More recent cameras have lower noise than older cameras.

If you are shooting at high ISO and only need 12 Mp or less, then the large pixel Sony A7Siii wins.

If you need high resolution, then high pixel count is the way to go.

Thank you very much for your detailed information!

Jahn

 JahnG's gear list:JahnG's gear list
Canon PowerShot A650 IS Panasonic ZS100 Pentax K100D Super Pentax K-r Pentax K-S1
JahnG
OP JahnG Veteran Member • Posts: 3,339
Re: Are bigger pixels less noisy?

Iliah Borg wrote:

JahnG wrote:

Iliah Borg wrote:

JahnG wrote:

how about the combined read noise of 50M pixels compared to the combined read noise of 20M pixels.

The combined read noise can be marginally, and I mean marginally, higher than the noise of a single pixel. Decreasing the maximum charge opens a possibility to improve performance in terms of noise.

Mr. Claff kindly keeps the original Aptina white paper on dual gain, it might help explaining the relationships: https://www.photonstophotos.net/Aptina/DR-Pix_WhitePaper.pdf

Thank you for your comments and for the Aptina white paper.

It's not just about Sony sensors, it's also about noise characteristics dependence on conversion gain and charge capacity.

Thanks, and yes.

Jahn

 JahnG's gear list:JahnG's gear list
Canon PowerShot A650 IS Panasonic ZS100 Pentax K100D Super Pentax K-r Pentax K-S1
Robert 23 Regular Member • Posts: 142
Re: Pixel size comparison & Aptina DCG

"If you are shooting at high ISO and only need 12 Mp or less, then the large pixel Sony A7Siii wins."
I suppose this is also valid for any 12 mpx FF sensor?

alanr0 Senior Member • Posts: 2,472
Quantum efficiency and signal to noise ratio
1

Robert 23 wrote:

"If you are shooting at high ISO and only need 12 Mp or less, then the large pixel Sony A7Siii wins."
I suppose this is also valid for any 12 mpx FF sensor?

Not necessarily.  Older cameras may not perform as well, and read noise is only part of the story.

The comparison in my post was purely on the basis of sensor read noise for the specific cameras I mentioned.

Recent dual conversion gain sensors from Sony have particularly low high ISO noise, typically around 1 e-

At ISO 6400 for the 12 Mp Sony 7Siii, Bill Claff report 0.9 e- rms.

At ISO 6400, the 61 Mp Sony 7Riv has input-referred read noise of 1.13 e-. Each pixel is 1/5.08 the areas of a 12 Mp sensor pixel. If we assume that noise adds in quadrature, the noise from a down-sampled Riv image will be 2.55 e- rms per 12 Mp equivalent pixel.

Compare this with the Nikon D3. An excellent 12 Mp camera in its day, but at ISO 6400 input-referred noise is 5.7 e-.

In spite of its larger pixels, the older Nikon D3 has twice the noise of the down-sampled 61 Mp Sony 7Riv

Quantum Efficiency

In fact, it is not even a clear win for the Sony 7Siii. The 7Riv has a higher quantum efficiency of 58%, compared with 51% for the 7Siii according to Bill's DxO derived data.

For an average of 80 photons per 12 Mp pixel, (around 1.4 % of highlight clipping at ISO 6400) the higher quantum efficiency of the 7Riv results in better signal to noise ratio than the 7Siii, in spite of lower read noise from the latter.

This is evident in one of Bill's photographic dynamic range plots, where his PDR measure accounts for differences in pixel count.

In contrast to engineering dynamic range, Bill's PDR does not emphasize the very deepest shadows, so the higher quantum efficiency of the 61 Mp Sony 7Rmk4 edges it slightly ahead of the 12 Mp 7Siii.

The 2007 Nikon 12 Mp Nikon D3 pre-dates the introduction of dual conversion gain, and lags by around a stop. Unsurprisingly, the 2004 12 Mp Nikon D2x can't compete with more recent sensors.

-- hide signature --

Alan Robinson

(unknown member) Forum Member • Posts: 61
Re: Quantum efficiency and signal to noise ratio
1

I don't know if that says anything but this is a test shot with my D700 (12,1mpx) at ISO 8000!! No editing. The lens is the old Nikon 28-105 AF-D f/3,5-4,5
Just RAW to jpeg and then an automatic noise reduction through "DeNoise Ai" software.
It looks to me like an iso 400 shot. Very clear and full of details.

Nikon D700-iso 8000

Iliah Borg Forum Pro • Posts: 28,665
Quantum efficiency, 93° ;)
1

Nick Zochios wrote:

I don't know if that says anything but this is a test shot with my D700 (12,1mpx) at ISO 8000!! No editing. The lens is the old Nikon 28-105 AF-D f/3,5-4,5
Just RAW to jpeg and then an automatic noise reduction through "DeNoise Ai" software.
It looks to me like an iso 400 shot. Very clear and full of details.

Nikon D700-iso 8000

Try comparing it to D4s or D6 without noise reduction

-- hide signature --
(unknown member) Forum Member • Posts: 61
Re: Quantum efficiency, 93° ;)
1

What is your point?

Iliah Borg Forum Pro • Posts: 28,665
Re: Quantum efficiency, 93° ;)
3

Nick Zochios wrote:

What is your point?

Try comparing D700 with noise reduction to D4s or D6 without noise reduction.

-- hide signature --
(unknown member) Forum Member • Posts: 61
Re: Quantum efficiency, 93° ;)
1

Ohhhhh I see.
Sorry, but I will destroy your dream.
It seems the modern software manages more or less to cancel the terms "Low light beast".
In other words, a 13 years old camera with the help of modern software gives at iso 8000 results which would normally give at iso 800.
I will come back and destroy your dream even further by showing you next time an edited shot at iso 12800 using Lr + DeNoise AI.

On the other hand, you are trying to compare a 300 bucks camera (D700) to cameras that cost thousands of bucks.
Are you willing to sell to me either your D4s or D6 for 300 bucks so that i don't need noise reduction software anymore?

Btw...This is a 12800 iso shot by a so-called " low light beast" camera (Nikon D750)
I smell the photographer didn't apply any noise reduction.
https://flic.kr/p/2k9ePjs
Please zoom in.
Would you call it a "low light beast"?
Does this shot need any noise reduction?
Would you call my D700 a "low light beast" comparing the D750 shot to this one (iso 12800 as well)
https://flic.kr/p/9bnhuF

J A C S
J A C S Forum Pro • Posts: 19,272
Re: Quantum efficiency, 93° ;)
6

Nick Zochios wrote:

Ohhhhh I see.
Sorry, but I will destroy your dream.
It seems the modern software manages more or less to cancel the terms "Low light beast".

It only seems that way to you.

In other words, a 13 years old camera with the help of modern software gives at iso 8000 results which would normally give at iso 800.

It does not. One example does not prove a statement; and what you posted lacks any texture to begin with.

(unknown member) Forum Member • Posts: 61
Re: Quantum efficiency, 93° ;)

What i posted is an iso 8000, without any editing, from an old technology camera which is has a usable limit of around 3200-4000 (the most)
Now at iso 8000, It has very little noise due to great software.
That's the fact. Either like it or not! (tbh i don't care

12800 and 25600 test shots from this "obsolete" camera are on their way..and i will keep you informed in order to make your new experienced comment.

Btw..do you have any comments on this shot? It comes from a low-light beast camera.
https://flic.kr/p/2k9ePjs

Keyboard shortcuts:
FForum MMy threads