22 replies to this topic

[sharkmelley]

Courtesy of a friend, I have managed get my hands on a modded Sony A7S (full frame sensor) for a few days to run a batch of tests and also compare it with my modded (APS-C) Canon 550D (Rebel T2i) and 350D (Rebel XT) cameras which I presently use for astrophotography.

 

The tests were performed in Airplane mode (i.e. battery draining Wi-Fi switched off) with all noise reduction switched off and RAW files created

 

As other reviewers have commented, there is a definitely a sharp drop in read noise between ISO 1600 and ISO 2000.  This can be clearly seen in Thierry Legault's graph: http://www.astrophot...s_measures.html

 

My measured read noise figures at ISO 1600, ISO 2000 were 4.2e RMS and 1.4e RMS  which are not much different to those obtained by Thierry and they certainly confirm this interesting behaviour. 

 

At ISO 2000 I measured the gain to be 2.1e-/ADU which means that unity gain is round about ISO 4000. 

 

For calculating relative RGB sensitivity of the cameras, a 2" Baader IR/UV filter was used in the imaging train.  For measuring relative H-alpha sensitivity, a 2" Baader H-alpha filter was used.  I used a uniformly lit computer monitor as the target.  It's not ideal but it has a standard brightness, unlike daylight.

The Canon 550D has always disappointed me because my measurements show that although the green channel collects 1.3x the number of photons per unit area than the 350D, the H-alpha sensitivity is identical.

My test show the Sony A7S collects 2.3x the number of photons per unit area than the 350D in the green channel and 2.0x in H-alpha.  This is a great result!

 

To get an idea of thermal noise requires the dark current to be measured.  Taking 2 hours of continuous 5min exposures at an indoor ambient temperature of 20C allows the sensor temperature to warm up and stabilise.  It is also typical of the type of imaging I perform - many hours of continuous imaging to extract the faintest detail. After 2 hours I measured the following dark currents:

Canon 350D (Rebel XT)   0.5e/pixel/sec

Canon 550D (Rebel T2i)   0.4e/pixel/sec

Sony A7S                          0.13e/pixel/sec

 

If the dark current is instead measured in terms of unit area then the Canon 550D is actually noisier than the 350D but the A7S is thermally much quieter, by a huge margin.

 

When a dark frame is examined, the pixel noise seen is mainly caused by the hot and warm pixels.  The following image shows how well dark frame subtraction removes this noise:

 

There are certainly some niggles with the Sony A7S:

- The Raw format is only 12 bit and to make things worse it is compressed (though the compression artefacts will probably never be seen at the optimum ISO of 2000-4000).  Why not an uncompressed 14 bit or 16 bit format?

- There is no integrated astro acquisition software available at present.  There probably never will be without using Wi-Fi control because that is the way Sony is heading.  However a combination of PHD Guiding, PHDMax and Sony Remote Camera Control means that at least the all important dithering can be done.

 

I think I'm prepared to work around these limitations to gain such a huge increase in performance:

- High quantum efficiency

- Extremely low read noise

- Very low thermal noise

 

Mark

Edited by sharkmelley, 29 March 2015 - 01:44 AM.

[axle01]

Awesome report Mark, I'm in Australia and I just had mine modified by Mmalik here on CN.

 

Only had one run with it last Sat night and very impressed with it, still a lot to learn though.

 

I purchased the Switronix battery pack for long sessions.

[axle01]

So Mark what should I turn off to save battery, things that I won't need for Astro photography.

 

This camera will only be used for Astro.

[whwang]

Hi,

 

Thank you for the report.

 

I am curious about the sharp drop of readout noise from ISO1600 to 2000.  sensorgen.info (which is based on DxOmark) doesn't show such a dramatic drop:

http://www.sensorgen.info/SonyA7S.html

 

Do you have insight on why this is the case?

 

Cheers,

Wei-Hao

[mmalik]

I am curious about the sharp drop of readout noise from ISO1600 to 2000.

 

Something I've been wondering as well; here... is another evidence that supports the sharp drop. Not so sharp drop in sensorgen plotted data here.... Regards

[sharkmelley]

I am curious about the sharp drop of readout noise from ISO1600 to 2000.  sensorgen.info (which is based on DxOmark) doesn't show such a dramatic drop:

http://www.sensorgen.info/SonyA7S.html

 

Do you have insight on why this is the case?

 

I'm no expert on the sensor technology but Bill Claff has investigated this point and explains it here: http://home.comcast...._Read_Noise.htm

 

Apparently it is because the sensor has 2 different conversion modes and it switches from one to the other between ISO 1600 and 2000.

 

I think that sensorgen uses a curve fitting approach to the DxOMark.  It might therefore tend to "smooth over" any steps it finds in the curve.  But I haven't got round to reading in depth about how exactly it works.

 

In any case, to obtain accurate measurements of noise in test images, it is far better to take pairs of shots and subtract one from the other (not forgetting to divide by square root of 2) - this removes any coherent sources of noise (e.g. warm and hot pixels).  This is what I've done in all my testing.

 

Mark

Edited by sharkmelley, 29 March 2015 - 01:53 AM.

[aacc66]

Courtesy of a friend, I have managed get my hands on a modded Sony A7S (full frame sensor) for a few days to run a batch of tests and also compare it with my modded (APS-C) Canon 550D (Rebel T2i) and 350D (Rebel XT) cameras which I presently use for astrophotography.

 

The tests were performed in Airplane mode (i.e. battery draining Wi-Fi switched off) with all noise reduction switched off and RAW files created

 

As other reviewers have commented, there is a definitely a sharp drop in read noise between ISO 1600 and ISO 2000.  This can be clearly seen in Thierry Legault's graph: http://www.astrophot...s_measures.html

 

My measured read noise figures at ISO 1600, ISO 2000 were 4.2e RMS and 1.4e RMS  which are not much different to those obtained by Thierry and they certainly confirm this interesting behaviour. 

 

At ISO 2000 I measured the gain to be 2.1e-/ADU which means that unity gain is round about ISO 4000. 

 

For calculating relative RGB sensitivity of the cameras, a 2" Baader IR/UV filter was used in the imaging train.  For measuring relative H-alpha sensitivity, a 2" Baader H-alpha filter was used.  I used a uniformly lit computer monitor as the target.  It's not ideal but it has a standard brightness, unlike daylight.

The Canon 550D has always disappointed me because my measurements show that although the green channel collects 1.3x the number of photons per unit area than the 350D, the H-alpha sensitivity is identical.

My test show the Sony A7S collects 2.3x the number of photons per unit area than the 350D in the green channel and 2.0x in H-alpha.  This is a great result!

 

To get an idea of thermal noise requires the dark current to be measured.  Taking 2 hours of continuous 5min exposures at an indoor ambient temperature of 20C allows the sensor temperature to warm up and stabilise.  It is also typical of the type of imaging I perform - many hours of continuous imaging to extract the faintest detail. After 2 hours I measured the following dark currents:

Canon 350D (Rebel XT)   0.5e/pixel/sec

Canon 550D (Rebel T2i)   0.4e/pixel/sec

Sony A7S                          0.13e/pixel/sec

 

If the dark current is instead measured in terms of unit area then the Canon 550D is actually noisier than the 350D but the A7S is thermally much quieter, by a huge margin.

 

When a dark frame is examined, the pixel noise seen is mainly caused by the hot and warm pixels.  The following image shows how well dark frame subtraction removes this noise:

SonyA7S_DarkFrames.jpg

 

There are certainly some niggles with the Sony A7S:

- The Raw format is only 12 bit and to make things worse it is compressed (though the compression artefacts will probably never be seen at the optimum ISO of 2000-4000).  Why not an uncompressed 14 bit or 16 bit format?

- There is no integrated astro acquisition software available at present.  There probably never will be without using Wi-Fi control because that is the way Sony is heading.  However a combination of PHD Guiding, PHDMax and Sony Remote Camera Control means that at least the all important dithering can be done.

 

I think I'm prepared to work around these limitations to gain such a huge increase in performance:

- High quantum efficiency

- Extremely low read noise

- Very low thermal noise

 

Mark

Mark,

 

I am dealing with the issue of dithering with my a7s (I am new to this business of DSLRs/Mirrorless cameras, so bear with me)... and found out that you probably need ~12 pixes of dithering with any DSLR (and I think this applies to any camera with a bayer color filter setup), but PHD Max only allow for a 4 pixel dither, which in my case it is on the QHY5L-II (3.75um pixels) and I using an OAG.  Is there a way to force PHDMax to dither more than their 4 pixels?

 

Thanks.

 

Alvaro

[sharkmelley]

The Sony A7S is not just any DSLR! 12 pixels of dithering?  I've never heard that suggested before - even for dealing with the infamous Canon banding problem (which the Sony definitely does not have).  I think that 4 pixels should work absolutely fine in your case, even though you are using a small pixel guide camera in OAG mode with a large pixel A7S sensor.  You just need to dither sufficently so that all the coloured pixels in the Bayer matrix are adequately sampled during the imaging run.

 

Mark

Edited by sharkmelley, 04 April 2015 - 02:05 AM.

[aacc66]

The Sony A7S is not just any DSLR! 12 pixels of dithering?  I've never heard that suggested before - even for dealing with the infamous Canon banding problem (which the Sony definitely does not have).  I think that 4 pixels should work absolutely fine in your case, even though you are using a small pixel guide camera in OAG mode with a large pixel A7S sensor.  You just need to dither sufficently so that all the coloured pixels in the Bayer matrix are adequately sampled during the imaging run.

 

Mark

I am hopeful from your comments and for sure I am going to give it a try. 

 

Thanks

 

Alvaro

[Samir Kharusi]

Not to put too much of a damper on what is obviously a disruptive new product, but the A7s has 8.5 micron pixels and is full 35mm format? The latter brings the usual headache of optics that deliver a flat enough image field. All camera superteles do that, but astro scopes often have issues. The 550D referred to above has 4.3 micron pixels, a pixel area that is a quarter of that for the A7s. Unless Sony has been horribly incompetent one would expect it to have a photon detection ability that is 4x better than the 550D, simple arithmetic. I.e the A7s should deliver images at an ISO 6400 to those delivered by the 550D at ISO 1600. So far so good. But now let's look at the optics required. OTAs invariably cost much more than the cameras. To sample the sky at, say, 1 arc-sec per pixel (seeing limit for long exposure DSOs at most sites on reasonably steady nights, the A7s needs a focal length of 1750mm, the 550D a much shorter 890mm. Consequently the A7s looks like a godsend for those with, e.g. SCTs with a reducer (still reasonably priced) but leads to ridiculously priced OTAs if one aspires to use a refractor. What I am attempting to say is it's still like it has always been; match your camera to your OTA... To keep overall Camera+OTA cost down we need to steer towards tiny pixels. Aperture costs much more than a new camera.

[aacc66]

Not to put too much of a damper on what is obviously a disruptive new product, but the A7s has 8.5 micron pixels and is full 35mm format? The latter brings the usual headache of optics that deliver a flat enough image field. All camera superteles do that, but astro scopes often have issues. The 550D referred to above has 4.3 micron pixels, a pixel area that is a quarter of that for the A7s. Unless Sony has been horribly incompetent one would expect it to have a photon detection ability that is 4x better than the 550D, simple arithmetic. I.e the A7s should deliver images at an ISO 6400 to those delivered by the 550D at ISO 1600. So far so good. But now let's look at the optics required. OTAs invariably cost much more than the cameras. To sample the sky at, say, 1 arc-sec per pixel (seeing limit for long exposure DSOs at most sites on reasonably steady nights, the A7s needs a focal length of 1750mm, the 550D a much shorter 890mm. Consequently the A7s looks like a godsend for those with, e.g. SCTs with a reducer (still reasonably priced) but leads to ridiculously priced OTAs if one aspires to use a refractor. What I am attempting to say is it's still like it has always been; match your camera to your OTA... To keep overall Camera+OTA cost down we need to steer towards tiny pixels. Aperture costs much more than a new camera.

 

I have an SCT and it has a focal length of 2321 mm... I have been using it for planetary work with some ok results, but I don't get often good seeing so I am dabbing on DSO work... therfore the A7s.

 

Alvaro

[sharkmelley]

Samir, that's a fair point you make about needing a good size corrected field to take full advantage of the full frame of the A7S.  However, I don't see whey it shouldn't be used on a shorter focal length scope, even though it will be undersampled.  I intend to use mine on a Tak Epsilon 180ED.  Dithering should take care of any colour sampling issues resulting from the Bayer matrix.

 

Mark

[jhayes_tucson]

"At ISO 2000 I measured the gain to be 2.1e-/ADU which means that unity gain is round about ISO 4000."

 

 

Uh, am I misunderstanding something here?  If the gain is 2.1e-/ADU at ISO 2000, shouldn't unity gain occur at a lower point--namely at ISO 1000?

 

 

John

[sharkmelley]

Uh, am I misunderstanding something here?  If the gain is 2.1e-/ADU at ISO 2000, shouldn't unity gain occur at a lower point--namely at ISO 1000?

 

 

John

 

No, at ISO 1000 the gain will be approx 4.2 e/ADU for the A7S.  As the ISO continues to reduce, it requires more and more photons to record a single digital unit.  This is the case for all digital cameras.  In normal daylight photography, if the scene you are photographing is too bright, then reducing the ISO is one of the things you might do to prevent the scene from saturating the sensor. 

 

Mark

Edited by sharkmelley, 05 April 2015 - 01:17 AM.

[ccs_hello]

John,

 

In astro world, be it CCD, DSLR, or webcam, for historical reasons, the gain is presented in e-/ADU unit, which really should be called reciprocal gain or inverse gain.

That higher the number, the less gain it is being applied.

 

Quite confusing isn't it?

 

The common-sense "gain" used everywhere else can be expressed as ADU/e-,

ISO (in some sense, it is a way to depict gain), and

others.

 

Clear Skies!

 

ccs_hello

[jhayes_tucson]

John,

 

In astro world, be it CCD, DSLR, or webcam, for historical reasons, the gain is presented in e-/ADU unit, which really should be called reciprocal gain or inverse gain.

That higher the number, the less gain it is being applied.

 

Quite confusing isn't it?

 

The common-sense "gain" used everywhere else can be expressed as ADU/e-,

ISO (in some sense, it is a way to depict gain), and

others.

 

Clear Skies!

 

ccs_hello

 

 

 

Thanks guys.  I'm certainly not an expert on detectors and I was thinking that amplifier output = sensor current * gain; where the amplifier output is passed through the ADU to produce a digitized signal, sensor current = photo electrons/time from the sensor = QE * incident photons/time, and gain = ISO setting.  If the amplifier produces 2e-/ADU at an ISO gain of 2000, then turning down the gain (i.e. ISO,) produces fewer photoelectrons per ADU.  So, yes it is confusing.

 

John

[calypsob]

CDs makes a cooled a7s body.. This camera amazes me.  I have read lots of great specs but really if you look at images shot with this camera it proves itself quickly. It picks up the ifn and reflection nebula around the Pleiades like its nothing in many images I have seen. There is a milkyway timelapse tutorial on you tube where a guy makes a hyper lapse with the a7s because he can shoot the milkyway at 1/30 of a second at ISO 400,000 or whatever it is in that area, and tbh the noise in the timelapse conversion is very minuscule. I want to make the switch to an a7s myself but need to prepare my gear for a ff sensor. By the time I do that I imagine something else will have come along. We shall see.its definitely got my vote.

[AnakChan]

I've got one of the CDS Astro A7s sitting in its box for the past month waiting for clear skies. I'm still a noob at this but if someone is willing to PM me on how to do those sharkmelley dark frames, I'll be happy to share the results here.

I just need to know the parameters of the test (any particular menu settings, any "warm up", also gather room temp suffices, etc.).

[mmalik]

CDs makes a cooled a7s body.. This camera amazes me...

 

 

I've got one of the CDS Astro A7s sitting in its box...

 

Calybsob, my concerns/questions here...

 

 

AnakChan, congrats on your a7S; do you now know exactly what kind of mod was performed to the sensor itself, I am referring to the sensor filter/s here? Regards

[AnakChan]

 

CDs makes a cooled a7s body.. This camera amazes me...

 

 

I've got one of the CDS Astro A7s sitting in its box...

 

Calybsob, my concerns/questions here...

 

 

AnakChan, congrats on your a7S; do you now know exactly what kind of mod was performed to the sensor itself, I am referring to the sensor filter/s here? Regards

 

 

From what I've been told by CDS, it's a full spectrum mod.The option after  that is if the buyer wants to have a clear or Hoya UV/IR filter in front of it. I went for the Hoya UV/IR filter.

[CCD1024]

Hello

I have since January a A7S which is modded (Astrodon filter)

 

I have the gap between 1600 and 2000 iso

 

I have taken an image at differents ISO with same other settings :

 

 

 

I have normalized (in fact an AUTO-STF in Pixinsight) to analyse each image in their Low light level

 

You can see the image looks the same details from 200000 iso to 2000 iso. Then, the effect of higher reaout noise for 1600 and 800 iso 

 

 

 

Cheers

[GJJim]

It looks like they did a great job integrating the cooler with this excellent camera. Too bad they decided to put everything in front of the sensor where it uses up precious back focus. A big advantage of mirrorless bodies is their short sensor distance and in this configuration CDS has eliminated the advantage.

[sharkmelley]

Hi CCD1024,

 

That's a very interesting test you have done there.  It demonstrates quite vividly the dramatic difference that takes place between ISO 2000 and 1600.  Also I notice horizontal streaking in the ISO 1600 and 800 which is absent at the higher ISOs.  The streaking is typical of what I get with my Canon cameras.

 

Thanks for sharing it

 

Mark