11 minutes ago, adamhanlon said:
Could the patterning be something to do with the way the camera debayers during processing? These ones in the image above are definitely color artifacts, which would prompt me to think this is a color issue rather than a phase detect one?
Banding etc. can all be caused in the algorithm too
The issue in my image is number 3 sensor reflections (of AF pixels) this is not the same as banding or striping and occurs in a specific scenario with all cameras. PDAF pixels only aggravate it
For what concerns the banding and striping
This thread explains the specific of Nikon Z7 https://petapixel.com/2018/10/01/nikon-z7s-banding-makes-it-fall-short-of-d850s-dynamic-range-report/
My understanding is as follows: AF pixels do not produce an image so the image has to be interpolated from near pixels. This is done before the data is de-bayered and saved as RAW. The process of replacing those pixels can create artefacts that in certain conditions lead to striping or banding.
If you increase the number of AF pixels and there is more interpolation to be done so smaller sensor like MFT or APSC are more prone to problems.
I have recently done a collaboration with Bill Claff the owner of the website Photons for Photos that has the largest collection of sensor data you can find on the internet. I submitted data for both my camera on the research for fixed pattern noise https://en.wikipedia.org/wiki/Fixed-pattern_noise#:~:text=Fixed-pattern noise (FPN),above the general background noise.
Fixed patterns means in short the chance you will see banding as it means the sensor is prone to create patterns in absence of light
I have submitted data for my two cameras that share the same sensor with the only difference being PDAF.
The measurement is conducted at low and high gain that is different for each manufacturer and camera.
So to give an example at ISO 200 which is low gain for both cameras the camera with contrast detect scored a value of 6% DSNU which is a measure of non uniformity so low values are good
The camera with PDAF scored 22% which is a fairly high value.
At high gain the camera with contrast detect scored 10% and the one with PDAF scored 14% which is a lower value of the low gain band.
In the specific case the AF pixels are 'hot' compared to the neighbour pixels at low gain when the high gain circuit kicks in the AF pixels remain as they are but the neighbour are also hot so the effect reduces and so does fixed pattern noise.
So as a paradox the impact is higher at low ISO and as this specific camera has high gain at 2000 it means really for all normal situations you have higher risk than at high gain. This is also confirmed by my experience I get sometimes pretty decent shots at ISO 4000 while I may have low quality shadows at ISO 400.
When you look at Nikon D850 vs Z7 the situation is almost similar at low gain the Z7 has 15% DSNU vs 8% of the D850 when they are both at high gain that kicks in pretty early at 400 the DSNU is the same. So this confirms what you read on petapixel this banding actually happens at low ISO which for most people is completely counterintuitive.
Nikon in addition has implemented a specific technique to minimise the striping that can actually create banding itself so if you look at Sony A7R series for example A7R3 you see that fixed pattern noise is lower an incredible 3% DSNU at low gain and dropping to 16% at high gain but here high gain is 640. From what I can see also canon dual pixel is effected as fixed pattern is noticeably worse in the EOS R than it was with traditional DSLR.
To what extent this is a problem is always the question and my view at current state of data is as follows:
1. Smaller sensor MFT/APSC with large number of AF pixels have pretty bad performance (Olympus OM-D EM1, Canon 90D, Fujifilm XT-3) in terms of fixed patter noise due to the density of AF pixels compared to the area. However the effect is stable or even reduced at high ISO
2. Larger sensors are also affected (Nikon Z, Canon R) to what extent it depends on the technology and interpolation but the trend is stable or getting worse at high ISO
3. Contrast detect sensors or traditional DLSR without AF pixels are generally better especially when high dynamic range is required at low ISO
The issue has not been investigated as a trend at this stage and it needs significant effort and data however I can quite clearly see the trend. When I submitted this problem to Bill Claff he wrote me
The PDAF pixels are different and certainly can cause more noise particularly Fixed Pattern Noise (FPN). I think there's a limit to how well any firmware can "fix" the effect.
I am not particularly interested in investigating the global effect of this issue however I do see a potential problem with underwater photography and banding as this is a problem that does occur at times in water. I also see an issue with dark shadow banding as that occurs a lot with flash photography. Historically banding has been linked to lack of bit depth however it seems much more interesting to look at fixed pattern noise as you can get banding also at high bit depth. When you don't get banding you get other effects generically described as 'noise in the shadows' but what really means is noise is not fine grain but clusters of patterns that look blotchy
I would be interested looking at low ISO shots from any of the candidate cameras to see to what extent this happens. I do not have any of those cameras nor am planning to buy them but that is unrelated to this issue.
