bclaff
Forum Pro
Note: HG and LG not HCG and LCG.
-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
My Panasonic S1 II tech thread
- Thread starter Horshack
- Start date
Noted. My bias tells me HG and LG are just shorthand for HCG+HG and LCG+no gain
The change notes describe a driver setting where additional analog gain in dB can be applied to Clear HDR's HG path. I don't see that setting reflected in the driver code yet, so it's not clear what chip register that driver setting translates to.
If Clear HDR is using analog gain only and the additional analog gain setting dB is a chip-level register value then it's not clear why it would be called "additional gain" instead of just "gain", irrespective of whether Clear HDR uses HCG or not for its base-level gain.
To see if it's visually obvious whether the HDR feature of the mechanical ISO 100 is using HCG or not here's a noise comparison I just did with a -8EV exposure boosted in ACR, comparing the S1 II's Electronic shutter at ISO 100, 640, and 800 to the Mechanical shutter at ISO 100. I also added in my DGO sim using the electronic shutter ISO 100+800 raws:
Animation: Electronic shutter ISO 100, 640, and 800 vs Mechanical Shutter ISO 100 and DGO Sim
The mechanical ISO 100 looks much better than electronic ISO 640, so if the HDR feature on the mechanical ISO 100 were using analog-only gain then it's clearly not a gain that's just +0.33EV above ISO 640. It's possible it's using an analog gain much higher than 640+0.33EV but there's no way for me to check that against the electronic shutter since the latter starts using HCG at ISO 800 and above.
In contrast, the mechanical ISO 100 looks very close to electronic ISO 800, implying that its noise performance is close to the HCG used for electronic ISO 800. However, while it's close, it is slightly worse than the HCG of electronic ISO 800. That could mean the mechanical ISO 100 is still using HCG but is blending in the LCG path enough to worsen it slightly. It could also mean mechanical ISO 100 is using analog-only gain but at a much higher dB than ISO 640+0.33EV.
Last edited:
bclaff
Forum Pro
I think it's important to distinguish between conversion gain, which is when a charge in electrons is converted to a voltage, and analog gain, which is when a voltage is amplified. So that's why I think your labels should be HG and LG without any "C"Noted. My bias tells me HG and LG are just shorthand for HCG+HG and LCG+no gain
The change notes describe a driver setting where additional analog gain in dB can be applied to Clear HDR's HG path. I don't see that setting reflected in the driver code yet, so it's not clear what chip register that driver setting translates to.
...
Understood. Based on the noise comparison in my reply to you it's hard to see how the S1 II's mechanical shutter is achieving its full advantage without the read-noise reducing benefit of HCG.I think it's important to distinguish between conversion gain, which is when a charge in electrons is converted to a voltage, and analog gain, which is when a voltage is amplified. So that's why I think your labels should be HG and LG without any "C"Noted. My bias tells me HG and LG are just shorthand for HCG+HG and LCG+no gain
The change notes describe a driver setting where additional analog gain in dB can be applied to Clear HDR's HG path. I don't see that setting reflected in the driver code yet, so it's not clear what chip register that driver setting translates to.
...
dcassat
Senior Member
Your discovery here seems to be a feature that Panasonic should be highlighting in marketing to photographers - why aren't they?
We get extra DR at 100 ISO and for landscape photographers, that's always a welcome thing. And, it appears to be a Panasonic exclusive! But I find it odd that you have to discover this in your evidentiary reverse-engineering rather than it being broadcast by the company itself.
Have you considered asking Panasonic? Or, perhaps you love the intellectual exercise?
Great (if not complex, sometimes above my head) reading!
We get extra DR at 100 ISO and for landscape photographers, that's always a welcome thing. And, it appears to be a Panasonic exclusive! But I find it odd that you have to discover this in your evidentiary reverse-engineering rather than it being broadcast by the company itself.
Have you considered asking Panasonic? Or, perhaps you love the intellectual exercise?
Great (if not complex, sometimes above my head) reading!
As described yesterday, I've set up a two LED configuration on my Arduino-based rolling shutter tester and modified the code to support frequencies faster than the usual 500 Hz I use.
Here is a photo of the setup:
I'm cycling between the two red and white LEDs at 2000 Hz, which means turning the red LED on for 250 us (while the white LED is off), then turning the white LED on for 250 us (while the red LED is off). I can cycle the LEDs faster but this is the fastest I can get clear demarcation of the two colors at the fastest shutter speed the S1 supports.
Here is a video frame of the S1 II 6k24p open-gate DR Boost-enabled capturing the cycling LEDs, both full-frame and a 200% crop:
Image: Two LEDs captured in video frame (full)
Image: Two LEDs in video frame (200% crop)
If the S1 II were performing any DOL dual-exposure, ie a T1/T2 long vs short exposure, then the colors should bleed across common rows. There is no bleeding outside of some slight smearing from the shutter speed, and the DR Boost-enabled capture looks the same as the DR Boost-disabled captured. except of course the latter having fewer bands due to the faster readout.
This should eliminate any remaining doubt about the camera using DOL for its HDR. The fact the HDR works with the mechanical shutter itself should've eliminated any doubt (outside of some hybrid electronic+mechanical readout scheme)...but I wanted to be certain.
Here is a photo of the setup:
I'm cycling between the two red and white LEDs at 2000 Hz, which means turning the red LED on for 250 us (while the white LED is off), then turning the white LED on for 250 us (while the red LED is off). I can cycle the LEDs faster but this is the fastest I can get clear demarcation of the two colors at the fastest shutter speed the S1 supports.
Here is a video frame of the S1 II 6k24p open-gate DR Boost-enabled capturing the cycling LEDs, both full-frame and a 200% crop:
Image: Two LEDs captured in video frame (full)
Image: Two LEDs in video frame (200% crop)
If the S1 II were performing any DOL dual-exposure, ie a T1/T2 long vs short exposure, then the colors should bleed across common rows. There is no bleeding outside of some slight smearing from the shutter speed, and the DR Boost-enabled capture looks the same as the DR Boost-disabled captured. except of course the latter having fewer bands due to the faster readout.
This should eliminate any remaining doubt about the camera using DOL for its HDR. The fact the HDR works with the mechanical shutter itself should've eliminated any doubt (outside of some hybrid electronic+mechanical readout scheme)...but I wanted to be certain.
Attachments
Last edited:
bclaff
Forum Pro
OK. Perhaps I missed something but you can't do two different conversion gains on the same pixel in the same image. So perhaps dual non-conversion gain with HCG also?Understood. Based on the noise comparison in my reply to you it's hard to see how the S1 II's mechanical shutter is achieving its full advantage without the read-noise reducing benefit of HCG.I think it's important to distinguish between conversion gain, which is when a charge in electrons is converted to a voltage, and analog gain, which is when a voltage is amplified. So that's why I think your labels should be HG and LG without any "C"Noted. My bias tells me HG and LG are just shorthand for HCG+HG and LCG+no gain
The change notes describe a driver setting where additional analog gain in dB can be applied to Clear HDR's HG path. I don't see that setting reflected in the driver code yet, so it's not clear what chip register that driver setting translates to.
...
I posed this very question on PST a few hours ago. Since that post I found two online resources that indicate both Samsung and SK hynix appear to have tech that can do this.OK. Perhaps I missed something but you can't do two different conversion gains on the same pixel in the same image. So perhaps dual non-conversion gain with HCG also?Understood. Based on the noise comparison in my reply to you it's hard to see how the S1 II's mechanical shutter is achieving its full advantage without the read-noise reducing benefit of HCG.I think it's important to distinguish between conversion gain, which is when a charge in electrons is converted to a voltage, and analog gain, which is when a voltage is amplified. So that's why I think your labels should be HG and LG without any "C"Noted. My bias tells me HG and LG are just shorthand for HCG+HG and LCG+no gain
The change notes describe a driver setting where additional analog gain in dB can be applied to Clear HDR's HG path. I don't see that setting reflected in the driver code yet, so it's not clear what chip register that driver setting translates to.
...
Last edited:
I've wrapped all my planned experiments. Here are the conclusions:
- The S1 II achieves its class-leading DR performance by utilizing a dual gain sensor readout with a single exposure, merging the two readouts into the synthesized raw. For the remainder of this post (and thread) I'll refer to this feature as DGO, "Dual Gain Output", even though that name might have different connotations in different contexts.
- Based on the both the analytical noise measurements and visual presentation I believe the S1 II's DGO is utilizing both LCG (low-conversion gain) and HCG (high-conversion gain), as opposed to just using different levels of analog gain for the two readouts. However, this detail can't currently be definitively proven.
- For still photography, DGO is currently only implemented when the mechanical shutter is used, either fully-mechanical or EFCS. It's presently not used for the electronic shutter. I believe this is not a technical limitation (since it's available for the video DR boost with an electronic shutter) but instead a decision by Panasonic due to the 2.3x slower sensor readout of DGO, which Panasonic probably thought was unacceptable for typical use. Panasonic has a history of favoring readout speed over IQ - for example, they use 12-bit readout instead of 14-bit for the shutterless S9.
- For video, DGO is utilized when the DR Boost feature is enabled.
- The FFTs show evidence of possible frequency filtering on the horizontal axis, although it's not yet clear if this is the consequence of the DGO merging algorithm or if it's noise filtering / reduction being applied. Based on the visual results I'm confident the S1 II's exceptional DR is not principally the result of NR.
- I'll be switching to other aspects of the camera to test but will return back to this if the situation warrants.
If my memory serves correctly I believe Richard Butler here at DPR has asked Panasonic in the past about such details and they weren't provided.
Maybe because it's against the will of "the deep state of camera industry" to introduce single-frame HDR to full frame mirrorless in 2025.
But Panny sold so little that no one cares. But they don't brag about it either.
As soon as they become mainstream, their cameras won't be unique anymore.
Last edited:
I've just published my comprehensive sensor readout speed measurements for the S1 II in my open-source Rolling Shutter project.
Summary for still imaging measurements:
https://horshack-dpreview.github.io/RollingShutter/
The landing page shows measurements for still imaging and common video modes, 8K/6K, 4K, and 1080.
The per-camera detail page has many more results, and for the S1 I measured nearly every video mode and configuration, including all the internal LGOP and ProRes Raw modes plus external BM raw. The detail page for the S1 II measurements is at:
https://horshack-dpreview.github.io/RollingShutter/#Panasonic S1 II
Summary for still imaging measurements:
- 14-bit full-sensor stills readout (6000x4000 3:2) is 1/69. This is the same as the Z6 III. This works out to 3.6us/row
- 12-bit full-sensor stills readout is 1/84, or 3us/row. This is slightly faster than the Z6 III's 12-bit readout of 1/90. Note that 12-readout is only used when shooting Burst SH mode with "speed priority".
- The S1 II uses 14-bit readouts for JPG vs 12-bit readouts for many competitors including Nikon, thus has the same readout speed as 14-bit raws.
- 6K 30fps Open Gate (5952x3968 3:2) LGOP is 1/69, same as the 14-bit stills readout. This demonstrates the S1 II is using 14-bit readout for video, as compared to 12-bit video readout on almost all other hybrids including the Z6 III. This results in higher DR but at the expense of slower readouts / more rolling shutter. For example, the Z6 III 12-bit 4k 30fps readout speed is 1/107 vs the 14-bit readout of 1/69 on the S1 II.
- Interesting, nearly all of the S1 II's 24fps measurements are slightly slower than their 30fps equivalents. For example, 6k 24fps Open Gate is 1/66.66 vs 1/69.38 for 30fps. This slower readout also applies to 48fps, which implies the speed difference is perhaps related to frame rate rounding of some type.
- DR Boost shows a 2.3x decrease in readout speed. For example, 6K 24fps Open Gate without DR Boost is 1/66.66 vs 1/29.27 with DR Boost.
- With a few curious exceptions, all of the oversampled video modes follow these same speeds after adjusting for row count differences, including from the aspect ratio crop. For example, 4k 30fps 3840x2160 16:9 30fps readout is 1/82. Note this is oversampled from 6000x3375 rather than 6000x4000, due to the 16:9 crop. This means each 4k 30fps readout is for 3375 rows instead of 4000, or 18.5% fewer rows than Open Gate. This is why the readout speed is 1/82 vs 1/69, or 18.5% faster.
- I also measured all of the APS-C (oversampled) and pixel-by-pixel 1:1 modes. Most produced expected results although many of the pixel readouts appear to be slower than expected for the number of 1:1 rows they're using, implying that the camera is reading out more rows than it's actually using for those modes.
- FHD (1920x1080) readouts are close to the full-sensor / oversampled rates for 6K/4K, implying the camera is doing full oversampling for HD, which is unusual for hybrids as most avoid this due to thermal issues. I'll need to compare the S1 II's FHD IQ vs competitors to confirm FHD oversampling.
- Internal ProRes Raw matched the readout speeds for LGOP after adjusting for row counts. However, external Blackmagic Raw (recorded over HDMI on a Blackmagic Video Assist 7" 12G) shows slightly faster readouts. Some of the DR boost enabled vs disabled readout speeds for BM raw show some discrepancies as well. The readouts aren't fast enough to be due to a drop down to 12-bit, so it's not clear why BM raw is show these differences.
https://horshack-dpreview.github.io/RollingShutter/
The landing page shows measurements for still imaging and common video modes, 8K/6K, 4K, and 1080.
The per-camera detail page has many more results, and for the S1 I measured nearly every video mode and configuration, including all the internal LGOP and ProRes Raw modes plus external BM raw. The detail page for the S1 II measurements is at:
https://horshack-dpreview.github.io/RollingShutter/#Panasonic S1 II
Last edited:
iso rivolta
Senior Member
- Messages
- 1,146
- Solutions
- 5
- Reaction score
- 848
Your investigation was spectacular and unveiled many details on how the S1II gains the extra DR.
Still, in my opinion, some smaller related questions remain unexplored, such as the above discussed linearity when grafting/blending the two gains and the effect on horizontal frequencies as shown by FFT. Do you think the latter could be an artifact derived from the direction on which the blended frame is generated (column by columns as opposed to row by row)?
I've actually done some preliminary work on this. I analyzed the SNR across the tonal range up to highlight clipping, looking for any drops suggestive of either a split continuous exposure integration (ie, T1/T2) or for SNR knees from blending. It wasn't an exhaustive check but I didn't find anything noteworthy.
As for the FFT, I'm still guessing it's demonstrating some type of spatial aspect to the blending algorithm used in the DGO but haven't spent much time coming up with possible algorithms that could produce that FFT result.
Last edited:
I was aware that SH Burst mode had an image vs speed priority setting, which determines whether the electronic shutter for SH Burst uses a 14-bit vs 12-bit readout.
In reviewing the manual today I discovered the H/H+ burst mode also has an image vs speed priority setting. But it doesn't affect the bit depth (always 14-bits). Instead, it determines whether the image has "wide dynamic range" or "standard dynamic range". That of course piqued my interest. When I checked the camera I found the setting is only available when the mechanical shutter is used. The instant I saw that I knew this setting could be used to control whether or not the camera employs DGO with the mechanical shutter in H/H+ mode.
Here's the manual page in question:
To verify my suspicion I took two ISO 100 blackframes in H burst mode - one with image priority and the other in speed priority. Again, both settings are with the mechanical shutter. Doing a cursory check of the noise stdev and FFTs, the "image priority" has the DGO characteristics and matches non-burst mode mechanical shutter. The "speed priority" blackframe does not have DGO characteristics and matches the electronic shutter in noise and FFT presentation.
For the H/H+ priority setting the manual page above even says "speed is prioritized more as the ISO sensitivity is increased", which matches my prior observations that the DGO benefit disappears at the HCG ISO since there is no LCG to merge with at those ISOs.
It's very strange how this is the only place in the entire manual that even obliquely references the fact the camera has a DR-improving feature for stills. Stranger still is how it's not referenced in any of Panasonic's marketing materials either.
In reviewing the manual today I discovered the H/H+ burst mode also has an image vs speed priority setting. But it doesn't affect the bit depth (always 14-bits). Instead, it determines whether the image has "wide dynamic range" or "standard dynamic range". That of course piqued my interest. When I checked the camera I found the setting is only available when the mechanical shutter is used. The instant I saw that I knew this setting could be used to control whether or not the camera employs DGO with the mechanical shutter in H/H+ mode.
Here's the manual page in question:
To verify my suspicion I took two ISO 100 blackframes in H burst mode - one with image priority and the other in speed priority. Again, both settings are with the mechanical shutter. Doing a cursory check of the noise stdev and FFTs, the "image priority" has the DGO characteristics and matches non-burst mode mechanical shutter. The "speed priority" blackframe does not have DGO characteristics and matches the electronic shutter in noise and FFT presentation.
For the H/H+ priority setting the manual page above even says "speed is prioritized more as the ISO sensitivity is increased", which matches my prior observations that the DGO benefit disappears at the HCG ISO since there is no LCG to merge with at those ISOs.
It's very strange how this is the only place in the entire manual that even obliquely references the fact the camera has a DR-improving feature for stills. Stranger still is how it's not referenced in any of Panasonic's marketing materials either.
Attachments
Last edited:
Interceptor121
Forum Pro
- Messages
- 11,791
- Solutions
- 8
- Reaction score
- 9,603
If I remember correctly in SH mode the camera only uses electronic shutter 12 or 14 bit depth wont make a difference to DR if the improvement is only available with mechanical shutter
in H/H+ you can use mechanical shutter instead
--
If you like my image I would appreciate if you follow me on social media
instagram http://instagram.com/interceptor121
My flickr sets http://www.flickr.com/photos/interceptor121/
Youtube channel http://www.youtube.com/interceptor121
Underwater Photo and Video Blog http://interceptor121.com
If you want to get in touch don't send me a PM rather contact me directly at my website/social media
patrol_taking_9j
Well-known member
Don't multiple FF cameras support single-frame HDR capture now? BT.2100 PQ for Canon FF and BT.2100 HLG for Nikon, not sure for the other systems.
Watching intensely as I am extremely interested in seeing how IMX410/820 is being leveraged by Panasonic and wonder if Nikon won't follow suit.
That's something hugely different. HLG or PQ are ready to use file forma for display on HDR screens. So you can use the wider DR of such screens compared to standard JPEG fils. But you don't get to the capabilities of RAW files.
What Panasonic does is enlarging the DR for RAW files.
Not correct - there is a drop in DR for 14-bit vs 12-bit for the electronic shutter:
Animation: S1 II SH Burst, Image Priority (14-bit) vs Speed Priority (12-bit), -7EV
Now that we've discovered H burst mode has an image priority vs speed priority setting that controls whether DGO is used, today I did measurements of the continuous shooting rate between the two settings, to see if the extra processing time for DGO with the mechanical shutter is the same as the extra readout time I measured for DR Boost Open Gate 6k video. I also measured image priority at various ISOs to see when the camera stops using DGO.
Methodology
Shoot a 4 second burst. Calculate the delta time between shots in each burst, with precision down to tens of milliseconds, using the following exiftool recipe that generates a comma-delimited file of shooting times with sub-second precision:
exiftool.exe -q -p "${CreateDate;$_=(split(' ',$_))[1]}.${SubSecTime}," path_to_files > creation_times.txt
Details:
Image Priority at ISOs <= 3200 add 20 ms of processing time to each shot. This exactly matches the increase in readout time for 6K Open Gate DR Boost video.
Interestingly, the camera stops using DGO at ISO 4000 - I expected it to stop at ISO 800, since that's the HCG point. If I compare the noise stdev for image priority at ISO 3200 (using DGO) vs speed priority at ISO 3200 (not using DGO), the speed priority actually has lower stdev. This indicates the exposure is actually noisier using DGO at ISO 3200 than not, which implies the firmware is selecting the wrong ISO for the optimal switching away point from DGO to non-DGO for image priority.
Methodology
Shoot a 4 second burst. Calculate the delta time between shots in each burst, with precision down to tens of milliseconds, using the following exiftool recipe that generates a comma-delimited file of shooting times with sub-second precision:
exiftool.exe -q -p "${CreateDate;$_=(split(' ',$_))[1]}.${SubSecTime}," path_to_files > creation_times.txt
Details:
- -q is "quiet mode", which suppresses the printing of banner info for each file
- -p indicates a format string will be used to specify the output.
- ${CreateDate;$_=(split(' ',$_))[1]}.${SubSecTime}, is a format string that extracts the creation date/time tag (ex: "2020:02:08 14:07:46"), then splits it into two fields by the space delimiter between them (perl expression) and keeps only the time portion by indexing the second field via the [1] subscript (result: "14:07:16"), then appends a '.' followed by extracting and appending the sub-second value, then finally appends a comma at the end since we're building a comma-delimited file. Example result: "14:07:16.34,". This value will be interpreted as time by Excel/LibreOffice Calc - the internal representation of time values in Excel/Calc is in seconds, represented as a decimal value between 0.00 and 0.99988426, so to get an integral number of seconds for the value multiply it by 86,400 (number of seconds in a day)
- Replace 'path_to_files' with the directory containing your images
- The output is a comma-delimited file named creation_times.txt.
Image Priority at ISOs <= 3200 add 20 ms of processing time to each shot. This exactly matches the increase in readout time for 6K Open Gate DR Boost video.
Interestingly, the camera stops using DGO at ISO 4000 - I expected it to stop at ISO 800, since that's the HCG point. If I compare the noise stdev for image priority at ISO 3200 (using DGO) vs speed priority at ISO 3200 (not using DGO), the speed priority actually has lower stdev. This indicates the exposure is actually noisier using DGO at ISO 3200 than not, which implies the firmware is selecting the wrong ISO for the optimal switching away point from DGO to non-DGO for image priority.
Attachments
Last edited:
Similar threads
Keyboard shortcuts
- f
- Forum