Big DSLR sensors are stitched together

[MarkS]

Did you know that lithography masks used for modern sensors are not big enough to cover a full-frame sensor, so the mask operation is done in two halves?  Nor did I but it has some very interesting implications.

I have wasted a huge amount of time trying to track down the cause of the vertical discontinuity in my master flat, seen here stretched:


Here I have colour saturated the flat:


There is horizontal colour banding on the left hand side which is absent from the right hand side.  Similar effects have been appearing on the left hand side of my images and are very difficult to fully remove.

At least I now know the cause!

Another nice example can be seen in this diagram "Figure 30. Nikon Df. Sensor stitching structure" at the bottom of this page:
http://www.rawdigger.com/howtouse/rawdigger-histograms-overexposure-shapes

Mark

[RobertM]

What a great piece of detective work, god only knows how you tracked that one down !

Does the issue 'flat' out or is the final image compromised ?

I must admit that I haven't noticed that on my flats yet but it's probably because I haven't looked hard enough.

Robert

[MarkS]

What a great piece of detective work, god only knows how you tracked that one down !

Does the issue 'flat' out or is the final image compromised ?

I must admit that I haven't noticed that on my flats yet but it's probably because I haven't looked hard enough.

Robert

I spent weeks trying to get to the bottom of this one.  Eventually I realised that every flat I took (on the Tak, C11 and lenses) showed the same features, as did all flats I have previously taken at all stages of the various mods I performed.  So it wasn't caused by my mod.

I eventually posted a question on Cloudy Nights and on DPReview and got similar replies from some very knowledgeable folk. 

The issue doesn't seem to completely flat out.  Here is an uncalibrated sigma stack that shows the horizontal banding (it's the Shark in Cepheus).  I've applied modest PixInsight DBE as a final step.


Unfortunately, not all the banding calibrates out. So I still don't know where that leaves me

Mark

[Carole]

How very annoying, and well done sussing it out.

Do you capture your DSLR data to the SD card or download it onto the laptop?

I am asking because do you recall a conversation some time ago where someone on the forums discovered that if you leave your SD card in the camera whilst downloading data to the computer then this would cause horizontal banding (I have the explanation filed away somewhere).  If you take the SD card out it greatly reduces it, I have tried this and it does appear to improve/alleviate the situation.  I am only mentioning this in case it gives you any thoughts on the horizontal banding though of course these issues might be completely different with the A7S.

Carole

[Mac]

The sensor is a sensitive dynamic device. Until each pixel is read (and digitised) its value can change. The primary way pixel values change is when photons hit them. However, other electronic disturbance can also affect them. If the circuitry is insufficiently de-coupled/isolated then the signals being generated during the process of writing the data (to memory, SD card, USB, or wherever) might affect the values of the pixels yet to be read, and give rise to odd effects...

But thats  different issue compared to noise being introduced whilst copying from sd card to pc.
I agree that until the data is pulled off of the chip and saved to the card then there is a possibility of corruption, but that corruption is not the fault of being saved to the sd card
that is the due process of reading and storing the data, which is a different cause. I found the original post on astroshed which dosent go in to any detail except one persons view that it might have been the slow cards and that was just a hunch.

#back on topic#

If these sensors are stitched together and you can reproduce these images then a flat and bias must be able to eliminate the artifacts, otherwise that would mean
that there is something fundamentally wrong with the way we have been processing images.

So at the moment we produce 4 files,
Bias    Technically should include all the readout noise present in EVERY image (including the split)
Flat     Photographs all the debris in the optical train
Dark   Photographs all the noise present
Light   All the juicy bits.

Now i know we dont produce a dark for our flat as it is technically not noisy (no long exposure noise, no heat noise ect) the only noise present is the bias,
is the bias file to blame and is it consistently noisy with the strip being present in each image? I know it should be there in every image but are the internal electronics
and file process doing anything?

Remeber Nikon cameras do not produce a RAW image it is always processed, Canons do produce a RAW image, does the sony fall under the nikon side of not producing a true raw image? and this is a software processing artifact not a true sensor artifact?

If it was a true sensor artifact then it should be processed out with our normal chain of processing, if its software (camera) related then does it appear in every image.
i.e the camera says oh look shortish image, process out the line as it will be seen, longer image na dont bother as it will be hidden by light data, so in our light frames there is no line as its already been incamera processed, but in the bias and or flats its there, which then introduce the banding to the final images.

Just a thought

Mac.

[MarkS]

Mac,

I agree, they ought to calibrate out but they don't.  So the investigation continues. 

The banding in the flat of a stitched sensor is just an interesting fact I discovered along the way. 

Mark

[Fay]

What would we do without you, Mark? You will get there in the end

Fay

[The Thing]

What would we do without you, Mark?

We'd all do our astrophotography in blissful ignorance!

[Mac]

 

[Fay]

 

[MarkS]

Here's a couple of interesting examples, both white balanced and extremely stretched to show the effect.

Firstly a flat taken with an opaque acrylic sheet over the telescope mouth pointing upwards to a cloudy sky in daylight:



Now, an hour's worth of data with the same opaque acrylic sheet over the telescope mouth pointing upwards to a cloudy sky at night:



So clearly those dark stripes will never calibrate out.  But what on earth is causing them and why do they appear in all my deep-sky images?

Here's the result of using the daylight flat to calibrate the nighttime one and stretching:



You can now clearly see the stripes I get in my images.  I'm running out of ideas.

Mark

[Carole]

That's pretty bad, I think you have no alternative but to either get in touch with the manufacturer with your results and see if you can get any help from them (swap a sensor or something), but maybe the next one could be even worse, or to attempt to process out the problem every time which will be really annoying.  However this is extremely stretched, are you going to stretch your image this much? 

I also wonder why this only seems to affect one side of the sensor. 

Also what is the cause of that circular ring in the daylight stretch flat and also the image of the uncalibrated Shark image?

The recent image you presented (Horsehead) did not show this artifact, however of course underneath it all you know it is there and it must restrict your processing to some extent.  Are you sure that it doesn't get processed out when you calibrate flats with image data?

Very annoying for an otherwise excellent camera.

Carole

[JohnP]

Mark,

Think Carole could be right - Any chance of passing this by someone at Sony?

The stripes are really hard to see in your horsehead head image - Can barely make out a hint of the top one - If I stretch it then yes its there but I guess in your search for perfection its annoying... Also looks like on your stacked image there is also bands starting to show at bottom left. Is there any chance you could get a fellow AS7 user (Robert maybe) to repeat the tests to see if he gets same results? Maybe its just your sensor..? I don't suppose there is a chance that its linked in anyway with your Ha mod of the camera? Just trying to think of a few things for how you can move forward with this. The litho mask you mentioned is also really annoying.. theres a real left/ right defining line in flats


John

[Fay]

the trouble with cameras and scopes etc, it is not a level playing field.  They are never all the same, some are better and worse than others

[MarkS]

Think Carole could be right - Any chance of passing this by someone at Sony?

The stripes are really hard to see in your horsehead head image - Can barely make out a hint of the top one - If I stretch it then yes its there but I guess in your search for perfection its annoying... Also looks like on your stacked image there is also bands starting to show at bottom left. Is there any chance you could get a fellow AS7 user (Robert maybe) to repeat the tests to see if he gets same results? Maybe its just your sensor..? I don't suppose there is a chance that its linked in anyway with your Ha mod of the camera? Just trying to think of a few things for how you can move forward with this. The litho mask you mentioned is also really annoying.. theres a real left/ right defining line in flats

The artifact from the sensor left/right stitching varies quite a lot. I know of a couple of examples of Sony A7S (Robert's included) that have no artifacts at all.  I know of another couple (mine included) with obvious artifacts.

I'm doing a few more tests to definitively check if those dark bands are being caused by some kind of light leak.

The Horsehead doesn't show the problem too much because I deliberately didn't stretch it as far as I wanted to.  I have a few other images (including the Shark) which are affected quite badly and are waiting for me to create a correction frame.  At least I now know a correction frame is possible by taking "flats" under a dark sky.

Mark

[The Thing]

Deep Sky Stacker has always had an option to include Dark Flats in the calibration but I had never read a convincing argument for bothering. Now I have one

[RobertM]

I do have similar bands in the upper left quadrant, but they seem to be much much fainter (with 20 frames and flattened with Pixinsight DBE).  There is also very faint evidence of a vertical line

It looks to me like these are at the limits of the lithographic sensor production process of this technology, could also be that the limited production runs also introduce tolerance variations.

Its a real pain but I'm glad you now seem to have a solution.

Robert

[MarkS]

I have now concluded that I can never calibrate these bands away.

You can find the full gory details in the Cloudy Nights thread I started:
http://www.cloudynights.com/topic/519941-sony-a7s-split-sensor-and-banding/

You will also see that I'm not the only A7S imager to hit this issue.

I now have lots of data that I'm never going to be able to properly process.  I am NOT amused.

Mark

[Fay]

Mark your Cloudy Nights thread was quite scary to read. thing is would Sony take any notice as you have voided the guarantee. do you think there is a big change you get this banding on all full frame sensors, with longer exposures?

[MarkS]

Mark your Cloudy Nights thread was quite scary to read. thing is would Sony take any notice as you have voided the guarantee. do you think there is a big change you get this banding on all full frame sensors, with longer exposures?

I wish I knew the answer to that question.  I am ready to junk the Sony A7S and to never buy another Sony again.  Trouble is, how can I be sure that I wouldn't get the same problems with a full frame Nikon (which will also use a "stitched" Sony sensor)

Mark

[Carole]

What a nightmare Mark, just read the CN thread.

It isn't even as if the banding is consistently the same in each exposure otherwise you could perhaps write some sort of banding removal calibration.  I presume the banding is on the lights as well as the flats?

Carole

[MarkS]

It isn't even as if the banding is consistently the same in each exposure otherwise you could perhaps write some sort of banding removal calibration.  I presume the banding is on the lights as well as the flats?

Yes, the banding appears in the lights as well as the flats.  The worst thing is that I've now proved that the width of the bands will vary from light frame to light frame as the light pollution levels change during the night.  So there is no hope of ever calibrating it out.

Mark

[The Thing]

Looks like it's APS-C from now on then.

[MarkS]

Looks like it's APS-C from now on then.

Canon 7D MkII is a very tempting proposition.

[RobertM]

OMG what an issue !!!  It will affect all Nikon FF sensors as well as Sony (and whoever else uses the same technology).  I wonder whether any Nikon users have noticed.  Does it change over time as some seem to infer on CN or do you think that's a Red-Herring ?

Did you use bias, darks as well as the flat frames ?

I'll do some similar tests with mine when I get time in the next day or two.

I feel really bad about this since it was me who introduced you to the camera in the first place !

Well the Nikon 810Da is supposed to be for astrophotography so it would be interesting to see if they have the same problem.

Robert

[Carole]

I am presuming this only affects the large format sensors as both my CCD cameras have Sony chips.

Carole

[MarkS]

OMG what an issue !!!  It will affect all Nikon FF sensors as well as Sony (and whoever else uses the same technology).  I wonder whether any Nikon users have noticed.  Does it change over time as some seem to infer on CN or do you think that's a Red-Herring ?

It's a red herring

Did you use bias, darks as well as the flat frames ?

There's no hint of banding in the bias or darks unless you do a 20min dark, for instance.

I feel really bad about this since it was me who introduced you to the camera in the first place !

No hard feelings!

I am presuming this only affects the large format sensors as both my CCD cameras have Sony chips.

Yes, only large sensors have "stitching".

Mark

[RobertM]

All your images with that camera have looked superb to me and I've never noticed any artifacts.  Is the issue showing up now or have you just managed to process it out before ?

Robert

[MarkS]

All your images with that camera have looked superb to me and I've never noticed any artifacts.  Is the issue showing up now or have you just managed to process it out before ?

The only images I've posted are the images where the banding is not too much of a problem.  The Cepheus Shark image (earlier in this thread) is one of 3 or 4 images waiting for a solution to the banding issue.  It's been a long road to understanding the problem but now I've finally demonstrated there is no easy solution, I really don't know what to do.

I was hoping that experimentation would lead to a potential solution. Unfortunately, experimentation has shown the problem to be intractable

Mark

[RobertM]

That's a real friggin "^"%$^%$ !!!

It's difficult to understand that it can't be calibrated out but if you can't find a solution then it's a bit of a dead end.  I know you're very determined to solve this but could you sell the camera as modded for infrared ?

Robert

[Canadian Roger]

Will this affect the full frame Nikon astrophotography camera - the D810a?

There was a review of this camera in the American magazine Sky and Telescope.  I haven't read the article yet, but I know a couple of people who have Nikon cameras are very much interested in the D180a.  They claim to feel "unclean" if they use a Canon camera.

Canadian Roger
Freezing in Canada

[MarkS]

I definitely want to know the answer to this.  I plan to outline a step by step methodology to show the problem, that anyone can use.  The problem can then be "crowd solved" by owners of various cameras.  I'll post a request on Cloudy Nights, DPReview, SGL and hopefully we can get some momentum on this problem.

Mark

[MarkS]

I've now written up the methodology for investigating the issue on other cameras on the Cloudy Nights thread:
http://www.cloudynights.com/topic/519941-sony-a7s-split-sensor-and-banding/?p=6993007

Mark

[Carole]

I've been thinking that not every-one who has one of these cameras seems to suffer with the banding, it seems some do some don't. 

So could be be just bad luck, or maybe those who say the don't just think they don't have a banding problem and perhaps don't analyse their images or stretch them sufficiently to notice.

Also I have had an idea.  It's not ideal but might overcome the worst of the problem, and certainly is not a solution, but just might get you out of a frustrating hole sometimes.  This will only work if you image both sides of the meridian.
If you took some of the data from a meridian flip, some of the areas with the banding you might be able to do a layer and use the flipped data on those areas as this will be using the other wise of the chip. 

It would be very annoying to have to resort to this, but if you are doing a meridian flip anyway.

Carole

[Mike]

Wouldn't that just move the banding to the opposite corner?

[Carole]

Wouldn't that just move the banding to the opposite corner?

It would Mike, but I was thinking if he took twice as much data, he could use the good side on each side of the Meridian i.e:

Basically take a normal image a), then take some more b), stack and process both.  flip b) and place a) over the top, and rub out the offending areas revealing the better data below.

Long winded but might help if the banding really spoils an image.

Carole

[MarkS]

Hi Carole,

That idea would definitely work - it would reduce the overall effect of the banding.  I'm also working on another idea which I need to think some more about.

Mark

[The Thing]

Wouldn't that just move the banding to the opposite corner?

It would Mike, but I was thinking if he took twice as much data, he could use the good side on each side of the Meridian i.e:

Basically take a normal image a), then take some more b), stack and process both.  flip b) and place a) over the top, and rub out the offending areas revealing the better data below.

Long winded but might help if the banding really spoils an image.

Carole

Or simply invert the camera halfway through the imaging run...

[RobertM]

What about isolating the banding by subtracting a flat from a camera where the problem is not showing so much ?  In exactly the same conditions you should be able to do that successfully.

It might be a bit more challenging to process but you could then apply a negative of the banding.  I'm not quite sure how much the banding depends on flux levels; that would be the limiting factor.

Robert

[MarkS]

What about isolating the banding by subtracting a flat from a camera where the problem is not showing so much ?  In exactly the same conditions you should be able to do that successfully.

It might be a bit more challenging to process but you could then apply a negative of the banding.  I'm not quite sure how much the banding depends on flux levels; that would be the limiting factor.

That's why I did those experiments at different exposure levels - to test out how stable the banding is:
https://drive.google.com/file/d/0B3Ky5pyZvsINTTQwNllXTXFpWm8/view?usp=sharing

Unfortunately the banding shifts quite a lot with exposure level i.e. with the level of sky fog.  So it won't work - it's too difficult to match the banding to the image frames.

I'm now thinking of using a set of 30sec exposures to render the brighter stars properly (without star eating) mixed with a set of long exposures.  With the long exposures the banding is less prominent and the very faint stars are "protected" from the star eater by the sky fog noise.

Some good news is that Sony engineers are actually looking seriously at the star eater problem.  Personally I don't think they will come out with a firmware fix for existing cameras but it might mean future cameras may have the option to disable it.

Mark