a7RII spatial filtering in various shutter modes

In another thread, I was asked:

tn1krr said:

I've looked around quite a bit in these threads, but one question seems not at least very clearly answered: what (if any) are the differences in 12-bit modes? I mean are for example following exposures equal or different assuming same lens/f-stop

1. 5 sec bulb exposure @ISO 100

2. 5 sec shot with continuous shutter mode @ISO 100

3. 5 sec single shot with silent shutter @ISO 100

4. 5 sec single shutter shot @ISO 100 with long exposure NR on. Would this be essentially same as any of above if I did manual dark frame subtraction.

All of above cause 12-bit RAW, but what are differences? Spatial filtering on bulb, but do we know if this applies to all/any of others.

Click to expand...

The answer follows. I would appreciate insight into the periodic components that only appear in some modes, particularly from Horshack, Prof Hank, Jack Hogan, and/or Bill Claff.

I received a request to look at what, if any, lowpass filtering the a7RII performs on dark-field images at ISO 100. Dark field images are mostly Gaussian noise at higher ISOs, although they can contain low-frequency energy do to pattern variations. A standard technique to remove the pattern variations is to subtract two dark frames and analyze the result. I have used that technique before, and will again, but I have found that it confuses people, and I didn't do it here.

The first test case is a 6 second exposure using silent shutter, single shot mode, long exposure noise reduction (LENR, aka dark frame subtraction). The camera timed the exposure:







There is no evidence of any low pass filtering. There is evidence of horizontal pattern noise in the form of a high component at extreme low frequency.

Next up, a 6 second bulb exposure in single shot mode with LENR off:







We have evidence of lowpass filtering in both the vertical and horizontal directions. In addition, there is a periodic effect in the horizontal direction. I don't know what causes that, although I've seen it before in some a7S dark field images.

Now, the same bulb exposure with LENR on:







Not much different, except that the periodic component in the horizontal direction is reduced. I take that to mean that the periodic component has a pattern component (not variable frame to frame), and a variable component, but I suppose it's possible the processing is different.

And lastly, a 6 second camera-timed exposure in continuous shutter mode:







There is no evidence of lowpass filtering in the vertical direction. The periodic component falls in modulation index as the frequency goes up, however.

I seriously doubt whether there is intentional, anisotropous, shutter mode dependent, filtering being done by the Sony engineers. Therefore I think we can look to the vertical frequency lines to see what the camera's design engineers deliberately put there. That would say that the silent shutter image and the continuous shutter images have no lowpass filtering added, while both of the bulb images do.

Jim

--

Jim, the A7rII doesn't support LENR with the silent shutter, so the the first graph is effectively without LENR (the description of the graph includes a reference to LENR but I'm not sure if you meant to say it was off/NA)

Horshack said:

Jim, the A7rII doesn't support LENR with the silent shutter, so the the first graph is effectively without LENR.

Click to expand...

I had LENR turned off for the silent shutter image. Did I screw up the caption? I see that I did. It should read:

"The first test case is a 6 second exposure using silent shutter, single shot mode, long exposure noise reduction (LENR, aka dark frame subtraction) turned off."

Nice catch, Horshack, and I apologize for the error.

Jim

JimKasson said:

Horshack said:

Jim, the A7rII doesn't support LENR with the silent shutter, so the the first graph is effectively without LENR.

Click to expand...

I had LENR turned off for the silent shutter image. Did I screw up the caption? I see that I did. It should read:

"The first test case is a 6 second exposure using silent shutter, single shot mode, long exposure noise reduction (LENR, aka dark frame subtraction) turned off."

Nice catch, Horshack, and I apologize for the error.

Jim

--
http://blog.kasson.com

Click to expand...

haha, not at all - I figured you just left out the "turned off" suffix (I edited my reply to that effect just before you posted your reply).

This is great data. I don't have any conclusions yet but maybe some areas of exploration. We know from your measurements of the A7rII data (and previously, A7s data) that the SS is considerably noisier than the EFC/mechanical, and moderately noisier than the other "12 bit" modes like continuous and bulb. It's telling that the SS is the noisiest and also the only "12 bit" mode that has no filtering applied based on your measurements - I wonder how much of this noise difference can be attributed to the lack of filtering to reduce the noise. For that I suppose we'd need more detail on what the filtering is actually doing.

Another question to ponder is what exactly the "12 bit" mode is. I've been using quotes around it because it certainly seems it's not just a difference of final raw encoding of 12 vs 14 bits but in fact a different operating mode of the ADC. And it's not clear if the difference in noise you're measuring can be attributed to an ADC running in 12 vs 14 bit mode (perhaps it can, just not clear to me). And within that different operating mode perhaps there are different submodes at play based on the different noise/filtering you're measuring. For example, I wonder if Sony is not using CDS for some of the these modes, especially the SS - that would explain not just the extra noise but also perhaps an increase in FPN like banding, which might also explain the asymmetric filtering you're seeing based on the mode. FPN/banding would have to be measured between the modes to see if this is in fact the case. Btw this question of CDS was first brought up in the context of video mode for CMOS sensors here.

JimKasson said:

In another thread, I was asked:

tn1krr said:

I've looked around quite a bit in these threads, but one question seems not at least very clearly answered: what (if any) are the differences in 12-bit modes? I mean are for example following exposures equal or different assuming same lens/f-stop

1. 5 sec bulb exposure @ISO 100

2. 5 sec shot with continuous shutter mode @ISO 100

3. 5 sec single shot with silent shutter @ISO 100

4. 5 sec single shutter shot @ISO 100 with long exposure NR on. Would this be essentially same as any of above if I did manual dark frame subtraction.

All of above cause 12-bit RAW, but what are differences? Spatial filtering on bulb, but do we know if this applies to all/any of others.

Click to expand...

The answer follows. I would appreciate insight into the periodic components that only appear in some modes, particularly from Horshack, Prof Hank, Jack Hogan, and/or Bill Claff.

I received a request to look at what, if any, lowpass filtering the a7RII performs on dark-field images at ISO 100. Dark field images are mostly Gaussian noise at higher ISOs, although they can contain low-frequency energy do to pattern variations. A standard technique to remove the pattern variations is to subtract two dark frames and analyze the result. I have used that technique before, and will again, but I have found that it confuses people, and I didn't do it here.

The first test case is a 6 second exposure using silent shutter, single shot mode, long exposure noise reduction (LENR, aka dark frame subtraction). The camera timed the exposure:



There is no evidence of any low pass filtering. There is evidence of horizontal pattern noise in the form of a high component at extreme low frequency.

Next up, a 6 second bulb exposure in single shot mode with LENR off:



We have evidence of lowpass filtering in both the vertical and horizontal directions. In addition, there is a periodic effect in the horizontal direction. I don't know what causes that, although I've seen it before in some a7S dark field images.

Now, the same bulb exposure with LENR on:



Not much different, except that the periodic component in the horizontal direction is reduced. I take that to mean that the periodic component has a pattern component (not variable frame to frame), and a variable component, but I suppose it's possible the processing is different.

And lastly, a 6 second camera-timed exposure in continuous shutter mode:



There is no evidence of lowpass filtering in the vertical direction. The periodic component falls in modulation index as the frequency goes up, however.

I seriously doubt whether there is intentional, anisotropous, shutter mode dependent, filtering being done by the Sony engineers. Therefore I think we can look to the vertical frequency lines to see what the camera's design engineers deliberately put there. That would say that the silent shutter image and the continuous shutter images have no lowpass filtering added, while both of the bulb images do.

Jim

Click to expand...

Brilliant, useful and educating at the same time. While I'm not into full blown astrophotography I imagine the filtering does not really help with milky way shots either and I'm after those in a few weeks. Staying off the bulb seem like safest thing to do.

I am curious (cause I don't have the technical knowledge to interpret your data): there are mixed reviews of the A7rii out, some love it, other's find the camera does not live up to it's initial hype. What can you tell from an engineering perspective? Is the camera as good as you had expected? Better? Worse?

Jim, an interesting question was posed to me in a private chat. What might the noise and filtering characteristics be when multiple, 12-bit inducing camera modes are used together. For example, the silent-shutter used in combination with the continuous shooting mode. You've already examined one combo mode in your initial test - bulb with and without LENR. Not asking for an experiment necessarily - just posing the question out loud