What causes stripe noise on a7iii?

Started Mar 1, 2018 | Discussions
Alantkh Regular Member • Posts: 342
What causes stripe noise on a7iii?
5

A thread on the noise pattern shown in one of the early samples on a7iii.

observations

1) it’s a column defect but it does not spread the entire height of the image

2) there is a overexposed region somewhere along the column

3) the lines are not evenly spaced

it don’t think it has anything to do with reflection or stray light on the phase detection pixels.

I am assuming the sensor uses a column adc and looks like the overexposed region forces the gain of the adc to be set to very Low. That will obviously impact the adc making it more sensitive to noise.

Not sure why the pattern does not extend the entire height ... maybe Sony updates the adc gain constantly and don’t stick to one value throughout the entire column

JimKasson
JimKasson Forum Pro • Posts: 32,682
Re: What causes stripe noise on a7iii?
4

Alantkh wrote:

A thread on the noise pattern shown in one of the early samples on a7iii.

observations

1) it’s a column defect but it does not spread the entire height of the image

2) there is a overexposed region somewhere along the column

3) the lines are not evenly spaced

it don’t think it has anything to do with reflection or stray light on the phase detection pixels.

I am assuming the sensor uses a column adc and looks like the overexposed region forces the gain of the adc to be set to very Low. That will obviously impact the adc making it more sensitive to noise.

Not sure why the pattern does not extend the entire height ... maybe Sony updates the adc gain constantly and don’t stick to one value throughout the entire column

By the "adc", I'm assuming that you mean the programmable gain amplifier (PGA) ahead of the ADC. The ADC in the a7RIII is a differential comparator with a ramp on one input that is used to gate a counter.

I know of no mechanism to change the PGA gain within an exposure.

I question your use of the word "obviously" above. Changing the PGA gain will result in changes to the mean levels of the ADC counts (or DNs, if you prefer) stored in the raw file.

The pattern of the lines is similar to, if not identical to, the pattern that Bill Claff discovered in the a9 when looking for a different effect of the PDAF pixels. This is suggestive, not dispositive, though.

Jim

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon Z7 Sony a7R IV Fujifilm GFX 100 Sony a9 II +1 more
OP Alantkh Regular Member • Posts: 342
Re: What causes stripe noise on a7iii?
1

JimKasson wrote:

Alantkh wrote:

A thread on the noise pattern shown in one of the early samples on a7iii.

observations

1) it’s a column defect but it does not spread the entire height of the image

2) there is a overexposed region somewhere along the column

3) the lines are not evenly spaced

it don’t think it has anything to do with reflection or stray light on the phase detection pixels.

I am assuming the sensor uses a column adc and looks like the overexposed region forces the gain of the adc to be set to very Low. That will obviously impact the adc making it more sensitive to noise.

Not sure why the pattern does not extend the entire height ... maybe Sony updates the adc gain constantly and don’t stick to one value throughout the entire column

By the "adc", I'm assuming that you mean the programmable gain amplifier (PGA) ahead of the ADC. The ADC in the a7RIII is a differential comparator with a ramp on one input that is used to gate a counter.

I know of no mechanism to change the PGA gain within an exposure.

I question your use of the word "obviously" above. Changing the PGA gain will result in changes to the mean levels of the ADC counts (or DNs, if you prefer) stored in the raw file.

The pattern of the lines is similar to, if not identical to, the pattern that Bill Claff discovered in the a9 when looking for a different effect of the PDAF pixels. This is suggestive, not dispositive, though.

Jim

Jim,

do u have links to technical documents that describe the a7riii sensor?

The gain of the adc is not solely decided by the pga. It can be also affected by the sensitivity of the photodiode or the size of the cap that converts the photo current to voltage. That is why I did not say pga specfically.

U don’t have to change the gain during the exposure, which can be as small as 100 usec, as the information is stored in capacitor. U have to finish it between exposures though. This gives u like 10 msec which is a pretty Long time when circuits nowadays can run in terms of 10s of psec.

Btw, my “obviously” refer to that IF the gain of the adc is set very Low, it will obviously impact the noise. This is just basic adc design. I am not saying obviously that Sony is setting the gain very Low, I don’t know that.

bclaff Forum Pro • Posts: 11,569
Re: What causes stripe noise on a7iii?
3

JimKasson wrote:

Alantkh wrote:

... maybe Sony updates the adc gain constantly and don’t stick to one value throughout the entire column

...

I know of no mechanism to change the PGA gain within an exposure.

...

FWIW, this is possible (but surely not happening here).

Changing the PGA on a line by line basis is how Magic Lantern implements their Dual ISO feature on Canon cameras.

Regards,

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Tons o Glass 0 Class
Tons o Glass 0 Class Contributing Member • Posts: 776
I still believe it's related to PDAF pixels.
18

Alantkh wrote:

A thread on the noise pattern shown in one of the early samples on a7iii.

observations

1) it’s a column defect but it does not spread the entire height of the image

In the examples provided, it doesn't span the entire height/width of the sensor, but that doesn't mean it can't -- it depends on the light source interacting with the PDAF-masked pixels. The rows of PDAF pixels span the entire height/width of the sensor, so if the conditions are right, you may get these artifacts everywhere. See below.

2) there is a overexposed region somewhere along the column

This is an interesting point. Maybe there's something to that.

If whatever is reading rows of pixels off the sensor is being clocked faster than it can perform with 100% accuracy, there could be some corrupted data. Wouldn't it be sad if Sony had to lower the burst rate to 7fps or something? But I feel like it would be random noise at that point. And since it occurs on the same rows of pixels, which are definitely in a pattern (see below), I don't think the fact that there is an overexposed region somewhere along the row is a cause unless Sony has an algorithm that references the overexposed beginning of the row when adjusting the PDAF pixels in it.

3) the lines are not evenly spaced

It's a pattern. I've stared at these problematic images for a little too long (I'm weird). More on this below. Sidenote: the pattern reminds me of when there's a bad ribbon cable on a broken laptop display.

it don’t think it has anything to do with reflection or stray light on the phase detection pixels.

I think it does. I believe it follows the structure of the sensor very very well.

I am assuming the sensor uses a column adc and looks like the overexposed region forces the gain of the adc to be set to very Low. That will obviously impact the adc making it more sensitive to noise.

I believe lines we see are based on data the sensor actually sees. I don't believe it's a glitch.

Not sure why the pattern does not extend the entire height ... maybe Sony updates the adc gain constantly and don’t stick to one value throughout the entire column

Bear with me for a bit longer. Here's an shot taken with my A7II (a camera known to have on-sensor PDAF) with no lens attached, using my phone's LED flashlight as a "flare". The image has been processed to make the lines of PDAF pixels more apparent, but they are visible as soon as you open up the raw file for processing.

If you were to take a peek at the RAW file at 100% in Capture One 10, you'll notice that these lines are actually dotted as opposed to solid, which could make sense if I were able to compare these lines to a super-macro image of the sensor.

Sweet PDAF lines on my A7II. There is a pattern, and in this case it's ALL OVER THE IMAGE instead of in just one spot. It would have to be one major crazy flare for this to happen with a lens attached, but... Stare at it long enough and you'll see a pattern, or just read below.

100% crop of the image above. Again, some processing has been done to make the rows of PDAF pixels more apparent. Note how the lines of PDAF pixels are actually dotted.

The A7II's pattern repeats every 210 pixels, and is simpler than the A7III's. Let "P" denote a PDAF pixel row, and let any number denote the same number of rows of normal pixels. The row pattern is one of the following, starting from the top of the image above, "reading" down:

  • P 13 P 41 P 27 P 13 P 27 P 13 P 27 P 41
  • PP 12 PP 40 PP 26 PP 12 PP 26 PP 12 PP 26 PP 40 (same as above, but with pairs of PDAF rows - I really don't know which is the case without a super-macro of the sensor).
-- hide signature --

Now for the A7III:

Rulers have been placed over lines seen on one of the problematic DPR sample gallery files. The red square denotes the repeating pattern. Any 420 pixel wide window will give you a repeating pattern.

The pattern repeats every 420 pixels (double that of the A7II), and is more complex than the pattern seen on my A7II. The pattern is defined as follows (same symbols as A7II pattern above):

  • P 5 P 17 P 11 P 11 P 17 P 11 P 5 P 11 P 11 P 11 P 17 P 11 P 5 P 11 P 11 P 17 P 5 P 11 P 17 P 5 P 17 P 5 P 11 P 11 P 11 P 17 P 5 P 11 P 11 P 11 P 5 P 17 P 5 P 17 P 11

The in-camera JPEG engine probably uses a less complex interpolation algorithm than does Capture One, which might help explain why the lines we see in DPR's sample gallery are more solid, while the ones seen in Capture One are dotted(?).

If this fixed pattern noise is indeed caused by stray light hitting the PDAF pixels or light leaking into pixels adjacent to the PDAF pixels somehow, the fact that the A7III sensor is BSI could make it easier for these things artifacts to show up -- a BSI sensor is more capable of capturing off-axis light afterall (it's one of its main advantages over normal CMOS sensor tech)!

 Tons o Glass 0 Class's gear list:Tons o Glass 0 Class's gear list
Sony Alpha NEX-3N A3000 Sony a7 II YI M1 Canon EOS 550D +2 more
iso rivolta Contributing Member • Posts: 770
Re: I still believe it's related to PDAF pixels.

Thank you for the detailed analysis. It's obvious that the banding is again related to the masked pixel rows on the sensor, very similar with the banding from many other cameras with such AF, including A6000, https://www.dpreview.com/forums/thread/3678989?page=3

What's interesting is that besides the main pattern (the obvious lines parallel with the sensor's length) there is a secondary finer pattern perpendicular on the first one. Maybe demosaicing has a contribution to the latter.

Hopefully contrast detection will get fastest in the future so that these compromises will not be necessary anymore. Nikon, Pentax (in live view) and Panasonic do not use PDAF off the main sensor in their high end models and this seems to be the reason.

Tristimulus Veteran Member • Posts: 7,597
Re: What causes stripe noise on a7iii?
1

One claim is that this can happen with fast lenses at large apertures when shooting against strong highlights. Not when using slower apertures. If so, not that much of a problem.

Guess this will affect very few real life images (unless often shooting against strong highlights). Very interesting issue, and one to tease out as soon as the A7III arrive here.

Jonneymendoza2 Senior Member • Posts: 1,048
Re: What causes stripe noise on a7iii?
3

no issues on the a7r3

Magnar W
Magnar W Veteran Member • Posts: 5,685
And what do you do about it?
1

I just posted a tread where I want people to judge a Quick Fix made with Photoshop. I don't think the stripes should cause much extra work for photographers who need clean 100% image size, since banding rarely occure - and when it occurs, it can easily be fixed with good results.

 Magnar W's gear list:Magnar W's gear list
Sony a7R III Sony a7 Zeiss Loxia 21mm F2.8 Zeiss Loxia 35 Zeiss Loxia 85mm F2.4 +2 more
JimKasson
JimKasson Forum Pro • Posts: 32,682
Re: What causes stripe noise on a7iii?

Alantkh wrote:

JimKasson wrote:

Alantkh wrote:

A thread on the noise pattern shown in one of the early samples on a7iii.

observations

1) it’s a column defect but it does not spread the entire height of the image

2) there is a overexposed region somewhere along the column

3) the lines are not evenly spaced

it don’t think it has anything to do with reflection or stray light on the phase detection pixels.

I am assuming the sensor uses a column adc and looks like the overexposed region forces the gain of the adc to be set to very Low. That will obviously impact the adc making it more sensitive to noise.

Not sure why the pattern does not extend the entire height ... maybe Sony updates the adc gain constantly and don’t stick to one value throughout the entire column

By the "adc", I'm assuming that you mean the programmable gain amplifier (PGA) ahead of the ADC. The ADC in the a7RIII is a differential comparator with a ramp on one input that is used to gate a counter.

I know of no mechanism to change the PGA gain within an exposure.

I question your use of the word "obviously" above. Changing the PGA gain will result in changes to the mean levels of the ADC counts (or DNs, if you prefer) stored in the raw file.

The pattern of the lines is similar to, if not identical to, the pattern that Bill Claff discovered in the a9 when looking for a different effect of the PDAF pixels. This is suggestive, not dispositive, though.

Jim,

do u have links to technical documents that describe the a7riii sensor?

No. I wish I did. What I know about the sensor I got from reverse engineering and reading Sony patents, so it's not super-precise.

The gain of the adc is not solely decided by the pga. It can be also affected by the sensitivity of the photodiode or the size of the cap that converts the photo current to voltage. That is why I did not say pga specfically.

Apparently, you are using the TLA adc differently than I use it. I view the ADC as the circuitry than converts analog voltage to a digital representation. I don't think the source follower is part of the ADC. I don't think the DR-Pix conversion gain switching is part of the ADC. I don't think the PGA is part of the ADC. I don't think the CFA is part of the ADC. I don't think the photodiode itself is part of the ADC.

U don’t have to change the gain during the exposure, which can be as small as 100 usec, as the information is stored in capacitor. U have to finish it between exposures though. This gives u like 10 msec which is a pretty Long time when circuits nowadays can run in terms of 10s of psec.

Btw, my “obviously” refer to that IF the gain of the adc is set very Low, it will obviously impact the noise.

If the conversion gain is lowered, the input-referred noise will rise, but the noise in DNs will go down. You buy that? Same with the PGA gain. You buy that? Is that obvious?

This is just basic adc design.

Again, I'm not sure of the precise point you're making. Can you be quantitative?

I am not saying obviously that Sony is setting the gain very Low, I don’t know that.

If the gain is changed during the readout, the raw channels would be non-linear wrt photon count. I don't see that.

Jim

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon Z7 Sony a7R IV Fujifilm GFX 100 Sony a9 II +1 more
JoFa New Member • Posts: 11
Re: I still believe it's related to PDAF pixels.
1

Nice analysis so far. Is it known if the PDAF pixels are actually arranged in that pattern? I haven‘t read into the math of PDAF but if that is the pattern they arrange the pixels to do so I will as this makes me curious. That periodicity of pattern reminds me somehow to interference patterns.

oncemoreintothebreach New Member • Posts: 18
Re: What causes stripe noise on a7iii?
6

Trollmannx wrote:

One claim is that this can happen with fast lenses at large apertures when shooting against strong highlights. Not when using slower apertures. If so, not that much of a problem.

You keep saying that, but for certain types of photography (e.g., natural light portraits and concerts) that's exactly how many people shoot the majority of the time.

Tristimulus Veteran Member • Posts: 7,597
Re: What causes stripe noise on a7iii?

oncemoreintothebreach wrote:

Trollmannx wrote:

One claim is that this can happen with fast lenses at large apertures when shooting against strong highlights. Not when using slower apertures. If so, not that much of a problem.

You keep saying that, but for certain types of photography (e.g., natural light portraits and concerts) that's exactly how many people shoot the majority of the time.

Do not see what is wrong with my claim (case if).

If not, well then you have described one out of several possible scenarios. Have excluded none of those in my statement.

OP Alantkh Regular Member • Posts: 342
Re: What causes stripe noise on a7iii?

Trollmannx wrote:

oncemoreintothebreach wrote:

Trollmannx wrote:

One claim is that this can happen with fast lenses at large apertures when shooting against strong highlights. Not when using slower apertures. If so, not that much of a problem.

You keep saying that, but for certain types of photography (e.g., natural light portraits and concerts) that's exactly how many people shoot the majority of the time.

Do not see what is wrong with my claim (case if).

If not, well then you have described one out of several possible scenarios. Have excluded none of those in my statement.

can we keep out the non-technical non-cause related discussion in this thread? thanks

OP Alantkh Regular Member • Posts: 342
Re: I still believe it's related to PDAF pixels.

Tons o Glass 0 Class wrote:

Now for the A7III:

Rulers have been placed over lines seen on one of the problematic DPR sample gallery files. The red square denotes the repeating pattern. Any 420 pixel wide window will give you a repeating pattern.

The pattern repeats every 420 pixels (double that of the A7II), and is more complex than the pattern seen on my A7II. The pattern is defined as follows (same symbols as A7II pattern above):

  • P 5 P 17 P 11 P 11 P 17 P 11 P 5 P 11 P 11 P 11 P 17 P 11 P 5 P 11 P 11 P 17 P 5 P 11 P 17 P 5 P 17 P 5 P 11 P 11 P 11 P 17 P 5 P 11 P 11 P 11 P 5 P 17 P 5 P 17 P 11

The in-camera JPEG engine probably uses a less complex interpolation algorithm than does Capture One, which might help explain why the lines we see in DPR's sample gallery are more solid, while the ones seen in Capture One are dotted(?).

If this fixed pattern noise is indeed caused by stray light hitting the PDAF pixels or light leaking into pixels adjacent to the PDAF pixels somehow, the fact that the A7III sensor is BSI could make it easier for these things artifacts to show up -- a BSI sensor is more capable of capturing off-axis light afterall (it's one of its main advantages over normal CMOS sensor tech)!

wow I trust your pattern measurement is accurate. Is the pattern really accurate down to the last pixel? If it is, it will remove a LOT of possible theories. I have no idea why such a complex pattern will form....

I don't think Sony will EVER design a phase detection based on such a weird pattern...

OP Alantkh Regular Member • Posts: 342
Re: What causes stripe noise on a7iii?

JimKasson wrote:

The gain of the adc is not solely decided by the pga. It can be also affected by the sensitivity of the photodiode or the size of the cap that converts the photo current to voltage. That is why I did not say pga specfically.

Apparently, you are using the TLA adc differently than I use it. I view the ADC as the circuitry than converts analog voltage to a digital representation. I don't think the source follower is part of the ADC. I don't think the DR-Pix conversion gain switching is part of the ADC. I don't think the PGA is part of the ADC. I don't think the CFA is part of the ADC. I don't think the photodiode itself is part of the ADC.

Yes, I define the Gain of the ADC as the gain of the analog (light) to digital conversion. It is easier as a lot of them behave the same way. Example, the sampling capacitor or the PGA almost behave exactly the same way except on controls the gain from the photocurrent to voltage while the other is voltage to voltage gain. Similarly the photodiode controls the light to current gain.

U don’t have to change the gain during the exposure, which can be as small as 100 usec, as the information is stored in capacitor. U have to finish it between exposures though. This gives u like 10 msec which is a pretty Long time when circuits nowadays can run in terms of 10s of psec.

Btw, my “obviously” refer to that IF the gain of the adc is set very Low, it will obviously impact the noise.

If the conversion gain is lowered, the input-referred noise will rise, but the noise in DNs will go down. You buy that? Same with the PGA gain. You buy that? Is that obvious?

This is just basic adc design.

Again, I'm not sure of the precise point you're making. Can you be quantitative?

In ADC process, we want to use as big as gain as possible and as soon as possible. The explanation is that once you amplify the signal, it is less susceptible to any noise that comes in. However, you cannot set the gain so high such that it saturates any portion of the ADC.

With such low noise ADC, I believe Sony will always try to maximize the gain with respect to the amount of light the sensor is reading.

I suspect that the overexposure on part of the columns impacts in the way how Sony sets the gain of the ADC such that we have column errors.

I am not saying obviously that Sony is setting the gain very Low, I don’t know that.

If the gain is changed during the readout, the raw channels would be non-linear wrt photon count. I don't see that.

Jim

-- hide signature --

Did you do an analysis with a image with different light intensity on various parts of the image?

Or is your analysis based on uniform images?

OP Alantkh Regular Member • Posts: 342
Re: What causes stripe noise on a7iii?

JimKasson wrote:

Alantkh wrote:

Jim,

do u have links to technical documents that describe the a7riii sensor?

No. I wish I did. What I know about the sensor I got from reverse engineering and reading Sony patents, so it's not super-precise.

Sony publishes papers at ISSCC every year. I am just wondering if anyone has read any Sony paper that is similar to A7iii... I only have access to ISSCC publications at work and this is not remotely related to my work.

I design gigabit optical transceivers not image sensors.

JimKasson
JimKasson Forum Pro • Posts: 32,682
Re: What causes stripe noise on a7iii?

Alantkh wrote:

JimKasson wrote:

The gain of the adc is not solely decided by the pga. It can be also affected by the sensitivity of the photodiode or the size of the cap that converts the photo current to voltage. That is why I did not say pga specfically.

Apparently, you are using the TLA adc differently than I use it. I view the ADC as the circuitry than converts analog voltage to a digital representation. I don't think the source follower is part of the ADC. I don't think the DR-Pix conversion gain switching is part of the ADC. I don't think the PGA is part of the ADC. I don't think the CFA is part of the ADC. I don't think the photodiode itself is part of the ADC.

Yes, I define the Gain of the ADC as the gain of the analog (light) to digital conversion. It is easier as a lot of them behave the same way. Example, the sampling capacitor or the PGA almost behave exactly the same way except on controls the gain from the photocurrent to voltage while the other is voltage to voltage gain. Similarly the photodiode controls the light to current gain.

U don’t have to change the gain during the exposure, which can be as small as 100 usec, as the information is stored in capacitor. U have to finish it between exposures though. This gives u like 10 msec which is a pretty Long time when circuits nowadays can run in terms of 10s of psec.

Btw, my “obviously” refer to that IF the gain of the adc is set very Low, it will obviously impact the noise.

If the conversion gain is lowered, the input-referred noise will rise, but the noise in DNs will go down. You buy that? Same with the PGA gain. You buy that? Is that obvious?

This is just basic adc design.

Again, I'm not sure of the precise point you're making. Can you be quantitative?

In ADC process, we want to use as big as gain as possible and as soon as possible. The explanation is that once you amplify the signal, it is less susceptible to any noise that comes in. However, you cannot set the gain so high such that it saturates any portion of the ADC.

With such low noise ADC, I believe Sony will always try to maximize the gain with respect to the amount of light the sensor is reading.

I suspect that the overexposure on part of the columns impacts in the way how Sony sets the gain of the ADC such that we have column errors.

I am not saying obviously that Sony is setting the gain very Low, I don’t know that.

If the gain is changed during the readout, the raw channels would be non-linear wrt photon count. I don't see that.

Jim

Did you do an analysis with a image with different light intensity on various parts of the image?

Or is your analysis based on uniform images?

You never answered my questions.

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon Z7 Sony a7R IV Fujifilm GFX 100 Sony a9 II +1 more
JimKasson
JimKasson Forum Pro • Posts: 32,682
Mostly in the G2 plane, not in G or R!
2

Alantkh wrote:

A thread on the noise pattern shown in one of the early samples on a7iii.

observations

1) it’s a column defect but it does not spread the entire height of the image

2) there is a overexposed region somewhere along the column

3) the lines are not evenly spaced

it don’t think it has anything to do with reflection or stray light on the phase detection pixels.

I am assuming the sensor uses a column adc and looks like the overexposed region forces the gain of the adc to be set to very Low. That will obviously impact the adc making it more sensitive to noise.

Not sure why the pattern does not extend the entire height ... maybe Sony updates the adc gain constantly and don’t stick to one value throughout the entire column

I just looked at a raw file and the striping is mostly in the G2 plane, with a little bit in the B plane (that I missed at first), and nothing in the R and G planes.

Images enhanced for visibility:

G2

B

R

G

Jim

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon Z7 Sony a7R IV Fujifilm GFX 100 Sony a9 II +1 more
JimKasson
JimKasson Forum Pro • Posts: 32,682
Re: What causes stripe noise on a7iii?
3

Alantkh wrote:

Did you do an analysis with a image with different light intensity on various parts of the image?

Yes.

Or is your analysis based on uniform images?

Not exclusively.

Jim

-- hide signature --
 JimKasson's gear list:JimKasson's gear list
Nikon Z7 Sony a7R IV Fujifilm GFX 100 Sony a9 II +1 more
Keyboard shortcuts:
FForum MMy threads