Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Sony A7S - Split sensor and banding

  • Please log in to reply
137 replies to this topic

#1 sharkmelley

sharkmelley

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1998
  • Joined: 19 Feb 2013

Posted 29 November 2015 - 08:33 AM

I've noticed some really weird behaviour on my Sony A7S sensor.

 

Here's a flat taken on my Tak Epsilon 180ED:

a7s_flat.jpg

 

Now here it is stretched:

a7s_flat_stretched.jpg

 

Notice the vertical discontinuity that occurs precisely halfway across the sensor?  This happens at all ISOs and occurs with lenses as well as telescopes.

 

Now here it is with colour saturation:

a7s_flat_saturated.jpg

 

Notice the wide horizontal banding on the left hand side of the sensor?  It's completely different in character to the right hand half. Again this happens at all ISOs and occurs with lenses as well as telescopes.

 

Here's an example image (the Shark Nebula in Cepheus) where I've stacked the uncalibrated dithered frames with sigma rejection and then applied PixInsight DBE to highlight the background features:

shark_uncalibrated.jpg

 

Again the vertical discontinuity and the banding on the left hand side is clear.  If I calibrate with flats then the vertical discontinuity is removed but I can never remove the banding.  This has become a fundamental limitiation to the amount of stretch I can apply to my images.

 

So what is going on with this sensor?  Is it using different circuitry on each half of the sensor, perhaps?

 

Mark

 

 

 



#2 David-LR

David-LR

    Explorer 1

  • -----
  • Posts: 82
  • Joined: 30 Nov 2013
  • Loc: AR

Posted 29 November 2015 - 09:28 AM

Does this behavior or the severity change with ISO settings?  It does look like the sensor is reading out to two different readout circuits. 



#3 AnakChan

AnakChan

    Viking 1

  • -----
  • Posts: 655
  • Joined: 01 Sep 2014

Posted 29 November 2015 - 10:22 AM

Funny, I don't see that with mine....granted it's a CDS A7s mod, but still a Sony A7s sensor :-

 

 

Attached Thumbnails

  • StretchedFlatISO3200.jpg


#4 sharkmelley

sharkmelley

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1998
  • Joined: 19 Feb 2013

Posted 29 November 2015 - 10:45 AM

Funny, I don't see that with mine....granted it's a CDS A7s mod, but still a Sony A7s sensor :-

 

I forgot to say that mine was a master flat: 16 flat frames at ISO 100 summed together.

 

Mike Malik is seeing some kind of split sensor effect in a dark frame:

http://www.cloudynig...dslr/?p=6915945

 

Mark



#5 AnakChan

AnakChan

    Viking 1

  • -----
  • Posts: 655
  • Joined: 01 Sep 2014

Posted 29 November 2015 - 11:03 AM

 

Funny, I don't see that with mine....granted it's a CDS A7s mod, but still a Sony A7s sensor :-

 

I forgot to say that mine was a master flat: 16 flat frames at ISO 100 summed together.

 

Mike Malik is seeing some kind of split sensor effect in a dark frame:

http://www.cloudynig...dslr/?p=6915945

 

Mark

 

OK got what you're saying here. Here's my stack of 20 @ ISO3200. Looks like I've got that line too. For a moment I was wondering if it was because of the way you guys had darkened the glass bezel to block out the IR light in the A7s housing, but since I've got that too and my sensor is outside, I guess it's inherent to the sensor itself :-

 

Attached Thumbnails

  • MasterFlat_ISO3200.jpg

Edited by AnakChan, 29 November 2015 - 11:05 AM.


#6 ccs_hello

ccs_hello

    Cosmos

  • *****
  • Posts: 9970
  • Joined: 03 Jul 2004

Posted 29 November 2015 - 11:53 AM

Mark,

 

It seems to be related to the known artifacts over full-frame sized image sensor manufacturing process.

 

FF sized sensor being large, has known to be exceeding the Nikon photo-lithography's lens' maximum "sweetspot" area coverage in the light-mask stepper.

(Yes, SONY fab uses Nikon lens based photo-litho stepper.  Canon fab uses Canon lenses.)

This causes a full frame sized image sensor being exposed in two steps (one left and one right.)  It is known as stitching.

It is very hard to maintain sub-micron precision when a mechanical component (microstepping stage) is involved.

Obviously there will be chip that will be more obvious than other due to manufacturing tolerance.  

See this post on the wafer picture: http://www.cloudynig...m/#entry6909037

 

Clear Skies!

 

ccs_hello


  • sharkmelley likes this

#7 sharkmelley

sharkmelley

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1998
  • Joined: 19 Feb 2013

Posted 29 November 2015 - 01:11 PM

ccs_hello,

 

Thanks so much for your reply.  I never realised that large sensors were stitched together.

 

There's another interesting diagram here, "Figure 30. Nikon Df. Sensor stitching structure", at the bottom of the page:

http://www.rawdigger...exposure-shapes

 

Mark



#8 ccs_hello

ccs_hello

    Cosmos

  • *****
  • Posts: 9970
  • Joined: 03 Jul 2004

Posted 29 November 2015 - 01:51 PM

Mark,

 

Yes, the Df one seems to be more obvious.

Df and D4 image sensor are using Renesas fab (and Nikon in house technologies) and the part number is NC81366W

which is different than SONY Exmor sensor used in A7 family.

 

The gist is that no one wants to redesign a special manufacturing process and invest a high million dollar new design for a relative low volume, niche market product: full frame size image sensor :).

 

Clear Skies!

 

ccs_hello



#9 sharkmelley

sharkmelley

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1998
  • Joined: 19 Feb 2013

Posted 14 December 2015 - 01:17 AM

The real problem to me is the stripes that appear in the left hand side of all my images after calibration.

 

Last night I put an opaque acrylic sheet over the scope, pointed the scope at a cloudy sky and took an hour of data.

 

After calibration, stacking and very severe stretching this is what I ended up with:

 

a7s_flat_artifact2.jpg

 

You'll see that the stripes match the stripes in the Shark image above.

 

These same stripes appear in all my processed images.  In other words, the stripes in all the flats I take do not match the stripes in all the deep-sky exposures I take.  What on earth is going on? Possibly some non-linearity in the camera's response?

 

Whatever the reason, I now have a method of isolating those stripe so at least I have the possibility of creating a synthetic "stripe removing" flat.

 

Mark


Edited by sharkmelley, 14 December 2015 - 01:20 AM.


#10 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 14 December 2015 - 01:08 PM

I'm not seeing it in the recent flats I gathered. This is an uncalibrated integration of 40x1.3 sec flats @ ISO3200 with the result debayered.

 

Screen Shot 2015-12-14 at 10.56.31 AM.jpg

 

Is this something that just started, or has it been there from the beginning? Maybe heavy use of the camera makes this start up? Mine is still only a few months old. exiftool shows I had a shutter count of ~9200 at the time of capturing these flats.



#11 sharkmelley

sharkmelley

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1998
  • Joined: 19 Feb 2013

Posted 14 December 2015 - 02:53 PM

Here is the definitive proof of why my flats are not calibrating out the stripes.

 

Here is the "flat" I obtained from 1 hour of imaging the background night sky with an opaque acrylic sheet - I've stretched it to highlight the "features":

 

deepskyflat.jpg

 

 

Here is a daylight flat taken with the same opaque acrylic sheet, stretched in an identical manner:

 

daylightflat.jpg

 

The banding (both horizontal and vertical) is just different :(

 

This is the weirdest problem I have come across in years of imaging.

 

And why do they both have a strange colour gradient left to right?

 

Mark


Edited by sharkmelley, 14 December 2015 - 03:57 PM.


#12 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 14 December 2015 - 03:13 PM

Here's my integration for comparison, completely untouched. It is not debayered. It's 40x1.3s flats @ iso 3200.

 

https://www.dropbox....so3200.fit?dl=0

 

Is it possible you have a light leak? Maybe invert the camera and re-run the test?

 

What does a really long dark frame look like?


Edited by schwim, 14 December 2015 - 03:15 PM.


#13 sharkmelley

sharkmelley

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1998
  • Joined: 19 Feb 2013

Posted 14 December 2015 - 03:55 PM

Here's my integration for comparison, completely untouched. It is not debayered. It's 40x1.3s flats @ iso 3200.

 

https://www.dropbox....so3200.fit?dl=0

 

Is it possible you have a light leak? Maybe invert the camera and re-run the test?

 

What does a really long dark frame look like?

 

Thanks for that - it's very interesting.  Yours is a totally well-behaved flat.  You have no full frame sensor stitching artefact whatsoever i.e. no vertical line halfway across the sensor.  You have no colour banding and no left/right colour gradient.

 

For me, it's only the light frames that show problems.  My dark frames show no features and my bias frames show no features.  . 

 

As for light leaks, I can't be 100% certain at this point.  However, the only way a light leak could cause those dark stripes is if the dark stripes are shadows. 

 

Mark


  • skyler likes this

#14 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 14 December 2015 - 04:08 PM

Have you tried taking flats against a computer screen or other flat source in a completely dark room? Maybe wrap the whole camera assembly up to the OTA in heavy fabric to rule extraneous light out?



#15 sharkmelley

sharkmelley

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1998
  • Joined: 19 Feb 2013

Posted 14 December 2015 - 05:36 PM

Have you tried taking flats against a computer screen or other flat source in a completely dark room? Maybe wrap the whole camera assembly up to the OTA in heavy fabric to rule extraneous light out?

 

That's exactly how I take my flats.  However, when deep sky imaging, I don't have the camera light shielded.  I'll try that next time.  But all my images have stripes in the same place whatever the orientation of the camera on the scope and whatever part of the sky I'm imaging.

 

The one thing that bugs me is those dark stripes on the flat produced by the background night sky are not perfectly horizontal.  That could indicate they don't originate inside the sensor electronics.

 

Mark


Edited by sharkmelley, 14 December 2015 - 05:43 PM.


#16 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 14 December 2015 - 06:11 PM

Yes, that does sound weird. What if you expose the camera without any optics attached? Same banding?

 

I got some gunk on my sensor a few weeks back. I couldn't use a puffer to get it off, so I had to go the wet swap route. I think it was oil from a machined adapter. It took ~8 of the swaps to get the thing clean. In every case up to the last I ended up with streaks that looked weird on flat out of focus images but not quite like yours. Any chance you cleaned your sensor this way?



#17 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 29 December 2015 - 03:07 AM

I may have caught the banding bug. :(

 

I just noticed this when trying to recover from my bad flats experience:

 

Screen Shot 2015-12-29 at 1.04.04 AM.jpg

 

Please excuse the crappy state of the image...

 

One thing that I can say was different when capturing the light frames is it was colder than usual, below 40F.


Edited by schwim, 29 December 2015 - 03:08 AM.


#18 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 29 December 2015 - 10:46 AM

The Canon Banding Reduction script in PixInsight seems to be able to remove it.



#19 sharkmelley

sharkmelley

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1998
  • Joined: 19 Feb 2013

Posted 30 December 2015 - 01:05 AM

That certainly looks like the same problem. Banding on the left hand half of the sensor which doesn't calibrate out.

I'm still doing experiments to fully understand why this is the case. It is possible that longer exposures may be required to get the skyfog well above the banding. But using bulb mode triggers the star eater.

To be continued ...

Mark

#20 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 30 December 2015 - 01:37 AM

I'll do the same.  I almost exclusively use bulb mode at this point. I have no idea if there is a star-eating problem - I can't count that high. :)



#21 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 30 December 2015 - 09:24 PM

Thinking about this... 

 

I wasn't seeing the banding problem previously, at least not that I was aware of. I changed my rig somewhat. It's a stretch, but maybe something about it is contributing and causing the banding. I added two things to the imaging train - an ONAG and an HEUIB-II filter. The other thing that changed is it's been quite cold here at night relative to normal - in the low 40s and 30s. I can't imagine any of these things are contributing, but they shouldn't be ruled out.



#22 Corsica

Corsica

    Vendor (Innovationsforesight)

  • *****
  • Vendors
  • Posts: 510
  • Joined: 01 Mar 2010
  • Loc: PA - USA

Posted 31 December 2015 - 12:36 AM

Often when flat frame processing fails it can be traced back to some kind of straight light/internal reflection issue.
 

I would suggest to do again the acrylic sheet test without the HEUIB-II filter this time.

Since it is close to the camera you may have some back and forth reflections between the chip and the filter surfaces.
 

Just a guess though.


Edited by Corsica, 31 December 2015 - 12:38 AM.


#23 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 01 January 2016 - 02:47 PM

Took another set of lights and flats, re-used dark and bias master. It persists. I don't see it in the darks.

 

The CanonBandingReduction script in PI seems to get the job done at least partly. I'll need to experiment more with it.

 

Screen Shot 2016-01-01 at 12.33.40 PM.jpg



#24 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 01 January 2016 - 05:49 PM

Often when flat frame processing fails it can be traced back to some kind of straight light/internal reflection issue.
 

I would suggest to do again the acrylic sheet test without the HEUIB-II filter this time.

Since it is close to the camera you may have some back and forth reflections between the chip and the filter surfaces.

 

Not sure. It seems to be happening to others on different optical trains.



#25 schwim

schwim

    Viking 1

  • *****
  • Posts: 598
  • Joined: 23 Oct 2014

Posted 04 January 2016 - 05:42 PM

I'm done capturing data for my recent M81-82 project. I have a total of 24 subs in this at 450s each taken across 3 separate nights. Flats, darks, and bias all applied. This banding & split is definitely not pretty.

 

Both images below were integrated and processed in PI and are prior to final noise reduction efforts. They've had crop, multiple DBE hits, BN, CC, SCNR and TDVDenoise. The intent here is to show two things: the problem as it exists for me, and the effect of the CanonBandingReduction script from PI. Default linked STF is applied to both.

 

Without the CanonBandingReduction script:

 

noscript.jpg

 

Note in the image above the right side of the sensor has a very wide "band". Could be this is from the local light pollution and DBE couldn't get it out. Makes me wonder what the interplay of light pollution is here, and whether a dark site might fare better.

 

With the CanonBandingReduction script applied twice before DBE was applied:

 

script.jpg

 

I'm thinking the scripted helped somewhat. In either case it ain't pretty. Very disappointing. I'm hoping it will be easy to hide this through clever stretching.

 

Maybe time to move to that CCD... Hrm.

 

 




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics