Did I reach the R3 sensor readout limits?

[fijnheid]

The other weekend I was taking some photos at a national celebration. There was a stand where kids could explode the lids off of test tubes with a vinegar/baking soda mixture. I was taking pictures as the test tubes went off.

This was of course a very fast thing, and most of my pictures were either before or just after the fact. However in one of my pictures, I captured the exact moment of the 'explosion'. It was taken at f/4, 1/2000 shutter speed, ISO 400, using electronic shutter and shooting RAW. It was shot outside in clear daylight.

Zooming in on the 'explosion' I noticed very strong vertical banding, something I'd never seen before on this camera. (I've taken over 100k photos with the R3 probably.)

100% crop below.



Did I manage to make the sensor readout visible?

 

[John Sheehy]

The other weekend I was taking some photos at a national celebration. There was a stand where kids could explode the lids off of test tubes with a vinegar/baking soda mixture. I was taking pictures as the test tubes went off.

This was of course a very fast thing, and most of my pictures were either before or just after the fact. However in one of my pictures, I captured the exact moment of the 'explosion'. It was taken at f/4, 1/2000 shutter speed, ISO 400, using electronic shutter and shooting RAW. It was shot outside in clear daylight.

Zooming in on the 'explosion' I noticed very strong vertical banding, something I'd never seen before on this camera. (I've taken over 100k photos with the R3 probably.)

100% crop below.

Did I manage to make the sensor readout visible?

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

 

[John Sheehy]

The other weekend I was taking some photos at a national celebration. There was a stand where kids could explode the lids off of test tubes with a vinegar/baking soda mixture. I was taking pictures as the test tubes went off.

This was of course a very fast thing, and most of my pictures were either before or just after the fact. However in one of my pictures, I captured the exact moment of the 'explosion'. It was taken at f/4, 1/2000 shutter speed, ISO 400, using electronic shutter and shooting RAW. It was shot outside in clear daylight.

Zooming in on the 'explosion' I noticed very strong vertical banding, something I'd never seen before on this camera. (I've taken over 100k photos with the R3 probably.)

100% crop below.

Did I manage to make the sensor readout visible?

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

The banding occurs at every 12.5 pixels, which would fit the 4000 pixel short side of the sensor 320 times, and the rolling shutter takes 4.8ms on the R3, so I figure the frequency of the light, if it is strobing, is about 66,667Hz.

Only a global shutter could hide that, with a very short exposure. Even a global shutter could show banding if something moved through the frame very fast with a strobing light source.

 

[fijnheid]

The other weekend I was taking some photos at a national celebration. There was a stand where kids could explode the lids off of test tubes with a vinegar/baking soda mixture. I was taking pictures as the test tubes went off.

This was of course a very fast thing, and most of my pictures were either before or just after the fact. However in one of my pictures, I captured the exact moment of the 'explosion'. It was taken at f/4, 1/2000 shutter speed, ISO 400, using electronic shutter and shooting RAW. It was shot outside in clear daylight.

Zooming in on the 'explosion' I noticed very strong vertical banding, something I'd never seen before on this camera. (I've taken over 100k photos with the R3 probably.)

100% crop below.

Did I manage to make the sensor readout visible?

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

Sorry, I forgot to mention that part... It was in portrait orientation!

However, I don't think that strobing light is a part of this. The only illumination was daylight, which does not strobe.

My hypothesis (based on this image) is that the exposure gets recorded in lots of thin horizontal bands of 12 pixels high simultaneously; and that per band, it scans from the bottom to the top (or top to bottom, I'm not quite sure).
Because the explosion size and/or brightness would have significantly changed during that 1/2000th second exposure, the scanning of those lines recorded a different image from the start to the end of the scan, leading to the visible lines/banding.

The reason only the explosion would be affected, is that it's the only thing in that image that would have changed significantly within 1/2000th of a second. Everything else was pretty much the same 1/2000th of a second later.

 

[Ephemeris]

The other weekend I was taking some photos at a national celebration. There was a stand where kids could explode the lids off of test tubes with a vinegar/baking soda mixture. I was taking pictures as the test tubes went off.

This was of course a very fast thing, and most of my pictures were either before or just after the fact. However in one of my pictures, I captured the exact moment of the 'explosion'. It was taken at f/4, 1/2000 shutter speed, ISO 400, using electronic shutter and shooting RAW. It was shot outside in clear daylight.

Zooming in on the 'explosion' I noticed very strong vertical banding, something I'd never seen before on this camera. (I've taken over 100k photos with the R3 probably.)

100% crop below.

Did I manage to make the sensor readout visible?

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

Sorry, I forgot to mention that part... It was in portrait orientation!

However, I don't think that strobing light is a part of this. The only illumination was daylight, which does not strobe.

My hypothesis (based on this image) is that the exposure gets recorded in lots of thin horizontal bands of 12 pixels high simultaneously; and that per band, it scans from the bottom to the top (or top to bottom, I'm not quite sure).
Because the explosion size and/or brightness would have significantly changed during that 1/2000th second exposure, the scanning of those lines recorded a different image from the start to the end of the scan, leading to the visible lines/banding.

The reason only the explosion would be affected, is that it's the only thing in that image that would have changed significantly within 1/2000th of a second. Everything else was pretty much the same 1/2000th of a second later.

Horshack has some good information on the number of rows read out sequentially. I forget if it is 12 but rings a bell.

He keeps a database which would help to show this but perhaps ask him he may have some information that would help.

 

[Ephemeris]

My Canon R3 sensor readout speed measurements

A few months ago I started an open source / crowd source project on GitHub to create a definitive repository of sensor readout speed measurements, for both stills and video. The project uses the built-in LED on an inexpensive Arduino board as a standardized light source with custom software that...

www.dpreview.com

 

[JohnSil]

The other weekend I was taking some photos at a national celebration. There was a stand where kids could explode the lids off of test tubes with a vinegar/baking soda mixture. I was taking pictures as the test tubes went off.

This was of course a very fast thing, and most of my pictures were either before or just after the fact. However in one of my pictures, I captured the exact moment of the 'explosion'. It was taken at f/4, 1/2000 shutter speed, ISO 400, using electronic shutter and shooting RAW. It was shot outside in clear daylight.

Zooming in on the 'explosion' I noticed very strong vertical banding, something I'd never seen before on this camera. (I've taken over 100k photos with the R3 probably.)

100% crop below.



Did I manage to make the sensor readout visible?

fijn, I'm absolutely certain you reached the limits of the R3 camera sensor, shutter and processor etc. You need to make sure you order the new R1 on the day it is announced an hope you don't render it useless in the first few days!

My advice is to call the Guinness Book of World Records and then recreate the shot. Certainly there is a world record in there somewhere!!! ;-)

John

 

[fijnheid]

The other weekend I was taking some photos at a national celebration. There was a stand where kids could explode the lids off of test tubes with a vinegar/baking soda mixture. I was taking pictures as the test tubes went off.

This was of course a very fast thing, and most of my pictures were either before or just after the fact. However in one of my pictures, I captured the exact moment of the 'explosion'. It was taken at f/4, 1/2000 shutter speed, ISO 400, using electronic shutter and shooting RAW. It was shot outside in clear daylight.

Zooming in on the 'explosion' I noticed very strong vertical banding, something I'd never seen before on this camera. (I've taken over 100k photos with the R3 probably.)

100% crop below.



Did I manage to make the sensor readout visible?

fijn, I'm absolutely certain you reached the limits of the R3 camera sensor, shutter and processor etc. You need to make sure you order the new R1 on the day it is announced an hope you don't render it useless in the first few days!

My advice is to call the Guinness Book of World Records and then recreate the shot. Certainly there is a world record in there somewhere!!! ;-)

John

Hahah. For me, if anything it just speaks to how good the electronic shutter already is, given that outside of this one image I've literally never seen anything like it in any of my previous photos.

 

[John Sheehy]

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

Sorry, I forgot to mention that part... It was in portrait orientation!

However, I don't think that strobing light is a part of this. The only illumination was daylight, which does not strobe.

Are you sure? It does look to me like there is extra light on the test tube.

My hypothesis (based on this image) is that the exposure gets recorded in lots of thin horizontal bands of 12 pixels high simultaneously; and that per band, it scans from the bottom to the top (or top to bottom, I'm not quite sure).

Typically top to bottom in the recorded image, which is bottom to top in the camera.

I have examined the spacing of the stripes, and they are about 12.4 pixels, not 12.000, so I don't think that this is the parallel readout. I wrote code that put blue lines every 36 columns of pixels from the mouse pointer, and lined it up with the leftmost clear strip edges in a 300% upsample of the image, and the strip edges get a little bit more to the right of the blue lines as I look from left to right.

Because the explosion size and/or brightness would have significantly changed during that 1/2000th second exposure, the scanning of those lines recorded a different image from the start to the end of the scan, leading to the visible lines/banding.

What I would expect, if that were the case, was that the exposure on either side of the seams would be the same on average, meaning that there would be lots of areas where you couldn't see the seam, and when you did see the seam, it could be brighter on the left, or, brighter on the right.
In your image, however, the seam always modulates pretty much the same, to the right of the area of interest, and only have the "isolated strip" effect at the very left, where there is a gradient in the cloud.

Are you absolutely certain that the test tube did not have illumination?

The reason only the explosion would be affected, is that it's the only thing in that image that would have changed significantly within 1/2000th of a second. Everything else was pretty much the same 1/2000th of a second later.

We still have the average period of 12.4 pixels, though, so I'm not convinced that 12-row parallel readout is the only thing going on.

 

[Horshack]

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

Sorry, I forgot to mention that part... It was in portrait orientation!

However, I don't think that strobing light is a part of this. The only illumination was daylight, which does not strobe.

Are you sure? It does look to me like there is extra light on the test tube.

My hypothesis (based on this image) is that the exposure gets recorded in lots of thin horizontal bands of 12 pixels high simultaneously; and that per band, it scans from the bottom to the top (or top to bottom, I'm not quite sure).

Typically top to bottom in the recorded image, which is bottom to top in the camera.

I have examined the spacing of the stripes, and they are about 12.4 pixels, not 12.000, so I don't think that this is the parallel readout. I wrote code that put blue lines every 36 columns of pixels from the mouse pointer, and lined it up with the leftmost clear strip edges in a 300% upsample of the image, and the strip edges get a little bit more to the right of the blue lines as I look from left to right.

Because the explosion size and/or brightness would have significantly changed during that 1/2000th second exposure, the scanning of those lines recorded a different image from the start to the end of the scan, leading to the visible lines/banding.

What I would expect, if that were the case, was that the exposure on either side of the seams would be the same on average, meaning that there would be lots of areas where you couldn't see the seam, and when you did see the seam, it could be brighter on the left, or, brighter on the right.
In your image, however, the seam always modulates pretty much the same, to the right of the area of interest, and only have the "isolated strip" effect at the very left, where there is a gradient in the cloud.

Are you absolutely certain that the test tube did not have illumination?

The reason only the explosion would be affected, is that it's the only thing in that image that would have changed significantly within 1/2000th of a second. Everything else was pretty much the same 1/2000th of a second later.

We still have the average period of 12.4 pixels, though, so I'm not convinced that 12-row parallel readout is the only thing going on.

It's the parallel readout. Each row readout is relatively slow (even on a stacked sensor - it only achieves fast, full-sensor speed by doing multiple rows in parallel); this slowness creates localized exposure or motion artifacts at fast shutter speeds.

This occurs on every stacked sensor design, and even some designs that do multi-row readouts without stacked memory (such as the R5). I see the OP's effect on every sensor readout measurement I've done on stacked sensors. Here's a magnified view of it on the Z9:



This is also an issue for HSS flash photography, for the same reasons:

https://www.dpreview.com/forums/post/67631119

 

[Ephemeris]

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

Sorry, I forgot to mention that part... It was in portrait orientation!

However, I don't think that strobing light is a part of this. The only illumination was daylight, which does not strobe.

Are you sure? It does look to me like there is extra light on the test tube.

My hypothesis (based on this image) is that the exposure gets recorded in lots of thin horizontal bands of 12 pixels high simultaneously; and that per band, it scans from the bottom to the top (or top to bottom, I'm not quite sure).

Typically top to bottom in the recorded image, which is bottom to top in the camera.

I have examined the spacing of the stripes, and they are about 12.4 pixels, not 12.000, so I don't think that this is the parallel readout. I wrote code that put blue lines every 36 columns of pixels from the mouse pointer, and lined it up with the leftmost clear strip edges in a 300% upsample of the image, and the strip edges get a little bit more to the right of the blue lines as I look from left to right.

Because the explosion size and/or brightness would have significantly changed during that 1/2000th second exposure, the scanning of those lines recorded a different image from the start to the end of the scan, leading to the visible lines/banding.

What I would expect, if that were the case, was that the exposure on either side of the seams would be the same on average, meaning that there would be lots of areas where you couldn't see the seam, and when you did see the seam, it could be brighter on the left, or, brighter on the right.
In your image, however, the seam always modulates pretty much the same, to the right of the area of interest, and only have the "isolated strip" effect at the very left, where there is a gradient in the cloud.

Are you absolutely certain that the test tube did not have illumination?

The reason only the explosion would be affected, is that it's the only thing in that image that would have changed significantly within 1/2000th of a second. Everything else was pretty much the same 1/2000th of a second later.

We still have the average period of 12.4 pixels, though, so I'm not convinced that 12-row parallel readout is the only thing going on.

It's the parallel readout. Each row readout is relatively slow (even on a stacked sensor - it only achieves fast, full-sensor speed by doing multiple rows in parallel); this slowness creates localized exposure or motion artifacts at fast shutter speeds.

This occurs on every stacked sensor design, and even some designs that do multi-row readouts without stacked memory (such as the R5). I see the OP's effect on every sensor readout measurement I've done on stacked sensors. Here's a magnified view of it on the Z9:



This is also an issue for HSS flash photography, for the same reasons:

https://www.dpreview.com/forums/post/67631119

Thanks Horshack. If we read say blocks of 12 rows at once is the localised exposure due to a change in light?

Is the reason they read 12 rows adjacent to each other, rather than further across the sensor a structure constraint?

 

[John Sheehy]

We still have the average period of 12.4 pixels, though, so I'm not convinced that 12-row parallel readout is the only thing going on.

It's the parallel readout.

Then why is the period 12.4 rows on average? There may be some effect from the parallel readout, but something else seems to be going on, especially the middle and right areas, where there is a modulation with a period of 12.4 that is independent of the shape and motion of any explosion. The first few bands on the left, however, look like the exposure changes abruptly in strips. The end of the test tube is glowing, as if some devcie were behind it to make it into a flashlight.

Each row readout is relatively slow (even on a stacked sensor - it only achieves fast, full-sensor speed by doing multiple rows in parallel); this slowness creates localized exposure or motion artifacts at fast shutter speeds.

Yes, but the elephant in the room is the fact that there is a period of 12.4.

This occurs on every stacked sensor design, and even some designs that do multi-row readouts without stacked memory (such as the R5). I see the OP's effect on every sensor readout measurement I've done on stacked sensors. Here's a magnified view of it on the Z9:

The "12.4" isn't going away, just because we know what parallel readout can do.

This is also an issue for HSS flash photography, for the same reasons:

https://www.dpreview.com/forums/post/67631119

 

[Horshack]

We still have the average period of 12.4 pixels, though, so I'm not convinced that 12-row parallel readout is the only thing going on.

It's the parallel readout.

Then why is the period 12.4 rows on average? There may be some effect from the parallel readout, but something else seems to be going on, especially the middle and right areas, where there is a modulation with a period of 12.4 that is independent of the shape and motion of any explosion. The first few bands on the left, however, look like the exposure changes abruptly in strips. The end of the test tube is glowing, as if some devcie were behind it to make it into a flashlight.

Each row readout is relatively slow (even on a stacked sensor - it only achieves fast, full-sensor speed by doing multiple rows in parallel); this slowness creates localized exposure or motion artifacts at fast shutter speeds.

Yes, but the elephant in the room is the fact that there is a period of 12.4.

This occurs on every stacked sensor design, and even some designs that do multi-row readouts without stacked memory (such as the R5). I see the OP's effect on every sensor readout measurement I've done on stacked sensors. Here's a magnified view of it on the Z9:

The "12.4" isn't going away, just because we know what parallel readout can do.

This is also an issue for HSS flash photography, for the same reasons:

https://www.dpreview.com/forums/post/67631119

Are you measuring 12.4 rows on the raw data or a debayered image? If the latter then you're potentially looking at a demosaicing or processing boundary artifact.

 

[Horshack]

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

Sorry, I forgot to mention that part... It was in portrait orientation!

However, I don't think that strobing light is a part of this. The only illumination was daylight, which does not strobe.

Are you sure? It does look to me like there is extra light on the test tube.

My hypothesis (based on this image) is that the exposure gets recorded in lots of thin horizontal bands of 12 pixels high simultaneously; and that per band, it scans from the bottom to the top (or top to bottom, I'm not quite sure).

Typically top to bottom in the recorded image, which is bottom to top in the camera.

I have examined the spacing of the stripes, and they are about 12.4 pixels, not 12.000, so I don't think that this is the parallel readout. I wrote code that put blue lines every 36 columns of pixels from the mouse pointer, and lined it up with the leftmost clear strip edges in a 300% upsample of the image, and the strip edges get a little bit more to the right of the blue lines as I look from left to right.

Because the explosion size and/or brightness would have significantly changed during that 1/2000th second exposure, the scanning of those lines recorded a different image from the start to the end of the scan, leading to the visible lines/banding.

What I would expect, if that were the case, was that the exposure on either side of the seams would be the same on average, meaning that there would be lots of areas where you couldn't see the seam, and when you did see the seam, it could be brighter on the left, or, brighter on the right.
In your image, however, the seam always modulates pretty much the same, to the right of the area of interest, and only have the "isolated strip" effect at the very left, where there is a gradient in the cloud.

Are you absolutely certain that the test tube did not have illumination?

The reason only the explosion would be affected, is that it's the only thing in that image that would have changed significantly within 1/2000th of a second. Everything else was pretty much the same 1/2000th of a second later.

We still have the average period of 12.4 pixels, though, so I'm not convinced that 12-row parallel readout is the only thing going on.

It's the parallel readout. Each row readout is relatively slow (even on a stacked sensor - it only achieves fast, full-sensor speed by doing multiple rows in parallel); this slowness creates localized exposure or motion artifacts at fast shutter speeds.

This occurs on every stacked sensor design, and even some designs that do multi-row readouts without stacked memory (such as the R5). I see the OP's effect on every sensor readout measurement I've done on stacked sensors. Here's a magnified view of it on the Z9:



This is also an issue for HSS flash photography, for the same reasons:

https://www.dpreview.com/forums/post/67631119

Thanks Horshack. If we read say blocks of 12 rows at once is the localised exposure due to a change in light?

Is the reason they read 12 rows adjacent to each other, rather than further across the sensor a structure constraint?

Yep, the intersection of the row readouts with the leading or trailing edge of a temporal or spatial light boundary.

 

[John Sheehy]

The "12.4" isn't going away, just because we know what parallel readout can do.

This is also an issue for HSS flash photography, for the same reasons:

https://www.dpreview.com/forums/post/67631119

Are you measuring 12.4 rows on the raw data or a debayered image? If the latter then you're potentially looking at a demosaicing or processing boundary artifact.

I forgot about lens corrections, and certainly that could make 12 into 12.4. However, I can't ignore how unnaturally bright the end of the test tube is, and how the periodicity in the middle and on the right looks like a duty cycle independent of subject matter.
--
Beware of correct answers to wrong questions.
John
http://www.pbase.com/image/55384958.jpg

 

[Ephemeris]

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

Sorry, I forgot to mention that part... It was in portrait orientation!

However, I don't think that strobing light is a part of this. The only illumination was daylight, which does not strobe.

Are you sure? It does look to me like there is extra light on the test tube.

My hypothesis (based on this image) is that the exposure gets recorded in lots of thin horizontal bands of 12 pixels high simultaneously; and that per band, it scans from the bottom to the top (or top to bottom, I'm not quite sure).

Typically top to bottom in the recorded image, which is bottom to top in the camera.

I have examined the spacing of the stripes, and they are about 12.4 pixels, not 12.000, so I don't think that this is the parallel readout. I wrote code that put blue lines every 36 columns of pixels from the mouse pointer, and lined it up with the leftmost clear strip edges in a 300% upsample of the image, and the strip edges get a little bit more to the right of the blue lines as I look from left to right.

Because the explosion size and/or brightness would have significantly changed during that 1/2000th second exposure, the scanning of those lines recorded a different image from the start to the end of the scan, leading to the visible lines/banding.

What I would expect, if that were the case, was that the exposure on either side of the seams would be the same on average, meaning that there would be lots of areas where you couldn't see the seam, and when you did see the seam, it could be brighter on the left, or, brighter on the right.
In your image, however, the seam always modulates pretty much the same, to the right of the area of interest, and only have the "isolated strip" effect at the very left, where there is a gradient in the cloud.

Are you absolutely certain that the test tube did not have illumination?

The reason only the explosion would be affected, is that it's the only thing in that image that would have changed significantly within 1/2000th of a second. Everything else was pretty much the same 1/2000th of a second later.

We still have the average period of 12.4 pixels, though, so I'm not convinced that 12-row parallel readout is the only thing going on.

It's the parallel readout. Each row readout is relatively slow (even on a stacked sensor - it only achieves fast, full-sensor speed by doing multiple rows in parallel); this slowness creates localized exposure or motion artifacts at fast shutter speeds.

This occurs on every stacked sensor design, and even some designs that do multi-row readouts without stacked memory (such as the R5). I see the OP's effect on every sensor readout measurement I've done on stacked sensors. Here's a magnified view of it on the Z9:



This is also an issue for HSS flash photography, for the same reasons:

https://www.dpreview.com/forums/post/67631119

Thanks Horshack. If we read say blocks of 12 rows at once is the localised exposure due to a change in light?

Is the reason they read 12 rows adjacent to each other, rather than further across the sensor a structure constraint?

Yep, the intersection of the row readouts with the leading or trailing edge of a temporal or spatial light boundary.

Thanks Horshack. Would a readout where the rows were not adjacent to each other have some improvements so that it wouldn't create bands?

 

[John Sheehy]

Thanks Horshack. Would a readout where the rows were not adjacent to each other have some improvements so that it wouldn't create bands?

No matter what order you read them, you're going to get motion artifacts. In sequential order, whether by line or strip, you have the simplest artifacts.

 

[Ephemeris]

Thanks Horshack. Would a readout where the rows were not adjacent to each other have some improvements so that it wouldn't create bands?

No matter what order you read them, you're going to get motion artifacts. In sequential order, whether by line or strip, you have the simplest artifacts.

--

The visuals will change in that you won't have effective wide bands / strips. If we spread them out across the sensor perhaps it's more acceptable to the human eye.

Beware of correct answers to wrong questions.
John
http://www.pbase.com/image/55384958.jpg

 

[fijnheid]

Which way was the sensor oriented? That can be very important. If it was in landscape orientation and you got vertical banding, it doesn't seem like it could be the e-shutter timing, but I can't think of anything for that scenario that would only affect the explosive cloud (vertical banding seems to be absent in the rest of the image). If it was in portrait orientation, then it could be that the artificial light illuminating the test tube strobed very fast and the e-shutter is too slow for that light source.

Faster rolling shutters get rid of 50/60Hz lighting issues, but it does not with much faster strobing in the hundreds or thousands of Hz, when the shutter speed is fast.

Sorry, I forgot to mention that part... It was in portrait orientation!

However, I don't think that strobing light is a part of this. The only illumination was daylight, which does not strobe.

Are you sure? It does look to me like there is extra light on the test tube.

Yes, I am sure. It was an open piece of pavement and grass, away from any of the stands because it needed some clear space for the flying lids. It was sunny and there was no lighting of any kind, even in the tents nearby.

Because the explosion size and/or brightness would have significantly changed during that 1/2000th second exposure, the scanning of those lines recorded a different image from the start to the end of the scan, leading to the visible lines/banding.

What I would expect, if that were the case, was that the exposure on either side of the seams would be the same on average, meaning that there would be lots of areas where you couldn't see the seam, and when you did see the seam, it could be brighter on the left, or, brighter on the right.
In your image, however, the seam always modulates pretty much the same, to the right of the area of interest, and only have the "isolated strip" effect at the very left, where there is a gradient in the cloud.

Are you absolutely certain that the test tube did not have illumination?

Strobing lights just cause a difference in the scene (brightness or color) during an exposure. However they are not the only way to cause a difference during an exposure. If anything significantly changes (spatially or in brightness or color) during an exposure, no matter the source, then that can have some visible effect on the readout.

The reason we see banding here could be because at the start of the exposure, the CO2/condensation cloud was smallish, and at the end, it might have been larger.

The brightness of the tube is just because of the way it was exposed, and because the lid was white. There was no other lighting than sunlight.

Here's larger crops from the "before", "during" and "after" shots. (Not sure if this was shot at 15 or 30 fps, but I guess 30.)

 

[John Sheehy]

Are you absolutely certain that the test tube did not have illumination?

Strobing lights just cause a difference in the scene (brightness or color) during an exposure. However they are not the only way to cause a difference during an exposure. If anything significantly changes (spatially or in brightness or color) during an exposure, no matter the source, then that can have some visible effect on the readout.

Yes, I know how that works, and I haven't denied that this may be at least part of the artifact, especially on the left side of the test tube, where it does like strips of different exposure times. However, a few things suggested to me a high-frequency strobe. When one is presenting an example of a sensor problem, the best thing is to supply a raw, or if not a raw, a conversion that is 1:1 with no lens corrections. The sRGB out put of RawDigger or FastRawViewer is uncorrected. If I saw a seam at exactly every 12 pixels, in the entire affected area, I would have been more likely to attribute it all to the 12-row parallel readout. What we actually have in the JPEG provided does not look like strips in the middle and the right parts of the affected area; it looks like a modulation from a duty cycle, possibly because of JPEG sharpening of the seams, combined geometric lens correction. The processing turned the edges of the strips into a mess.