#21  
Old 07-12-2019, 05:26 PM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
The main issue I am grappling with is that the noise is significantly greater below 120 gain (Roughly unity) but from 120 where the noise is greatly improved, the useable range of the sensor is down by two stops as the ADC will max out in value at roughly 16ke instead of 65.
Reply With Quote
  #22  
Old 07-12-2019, 07:04 PM
Atmos's Avatar
Atmos (Colin)
Ultimate Noob

Atmos is offline
 
Join Date: Aug 2011
Location: Melbourne
Posts: 6,430
As you increase the gain you decrease the read noise with CMOS sensors. This means that for the same exposure time you’ll get less noisy images... because the read noise is lower.

You used the example of 30s images in previous posts, at low gain you may not be getting even close to being read noise limited which will severely damage low signal areas. With 30s exposures at unity, the read noise is lower and that may not be an issue. This just means that with lower gains you need longer exposures to get above the noise floor.

Also, being below unity never means that you’re losing dynamic range. If you have a well depth of 60k with a 14-bit ADC you’re not losing dynamic range. The ADU is an Analog to Digital Converter as you’re well aware from radio applications, all you’re doing is changing it from a physical to a digital format. You’re never losing the range by cramming it from 60k to 16k simply because all of the forms of noise are scaled down as well.

This gets back to what I was saying a few posts back about what dynamic range is, it’s the amount of discrete bits. If your camera only has 12 bits worth of dynamic range ON SENSOR then whether it is a 14-bit or 16-bit ADC doesn’t matter because the other 2-4 bits is just noise error and padding.
Reply With Quote
  #23  
Old 08-12-2019, 08:20 AM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
All of what I am talking about does rely on one thing that I have not seen written explicitly but that is implied, that the sensor is 65Ke full well regardless of gain, that is that the gain is applied to the read output not to the capture. Given the gain is expressed in terms of electrons per ADU I think it reasonable.

Assuming that, at the gain that gives the best noise profile (120, just over unity for the ASI294) you have 0-65Ke range on the sensor, which needs 16 bits to quantise, but a 14 bit ADC. Examination of subs suggests that at 120 gain, data is truncated at 16Ke, the limit of 14 bits. I found some features in M42 that could be readily identified on both 30 and 300 second unstretched subs. At 30 seconds the .fit file it showed a brightness level of around 12K, divide that by four to get back to the 14 bit level and it is around 3K off the sensor. Multiply that by 10 to get a reasonable guestimate of what the sensor would read for the same feature at 300 seconds and it would be about 30Ke, about half of the sensors range. In the .fit file for 300 seconds it is topped out at max value. Rough working would suggest that particular feature should saturate the sensor after about 600 seconds (a bit less due to dark current) but with 16K steps in the ADC output, it would saturate in the output file somewhere around 150 seconds.

Rather than being a loss of dynamic range, the 14 bit conversion to me is a loss of potential dynamic range that would be very useful for astro imaging, you should be able to capture more, and stretch less.
Reply With Quote
  #24  
Old 08-12-2019, 09:39 AM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
Anyway, it is all fairly esoteric given the IMX294 sensor has a 14 bit converter and always will have. But i am prone to through experiments on the "What would happen if" variety.
Reply With Quote
  #25  
Old 08-12-2019, 11:49 AM
Atmos's Avatar
Atmos (Colin)
Ultimate Noob

Atmos is offline
 
Join Date: Aug 2011
Location: Melbourne
Posts: 6,430
https://astronomy-imaging-camera.com...94mc-pro-color

I checked out the camera on the ZWO website (link above) and it appears to have a interesting characteristic that is seen on a few CMOS sensors (not many at this stage) and that is a dual gain mode; this kicks in at gain 117 on your camera.
This hints at what you’re describing and is inherent to your camera but not cameras in general. As gain is increased the full well capacity (FWC) decreases as does read noise. Generally read noise drops slower than the decrease in FWC which is why the dynamic range decreases as gain increases. Your sensor though has an obscene drop in read noise around unity gain which recovers a LOT of dynamic range. It’s dropping from 6e- to 1.75e- instantly so you have no reason but to use unity gain because there is no loss.
Reply With Quote
  #26  
Old 09-12-2019, 08:36 AM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
Without seeing specs on the next generation of sensors it is hard to draw too many conclusions for anything but the IMX294 based cameras or others which may have some form of dual gain mode that creates a big drop in read noise. I have no idea whatsoever what the difference is at gain 120 that cuts the noise from approx 8 to approx 2 electrons but per my posts above, the difference in images shot is very noticeable and I will take it gladly.

As I have said previously, you can't and I don't blame Sony for putting in a 14 bit converter as the sensor is designed for a different application. It is hard to see a security camera needing to provide a faithful reproduction of a scene where real world sensor levels you want to capture and quantise can literally span the range of two to 65K electrons in a single exposure of 300 seconds or more, just that it would be a great party trick for an astro cam to be able to do. The party trick of this sensor as a security cam is concurrent dual length exposures for HDR output. The RGGB pixel arrangement is actually RRRR-GGGG-GGGG-BBBB as each pixel is a block of four sub pixels of the same colour which can be exposed for different times in one shot.

Another interesting but significantly pointless point. There does not seem to be any such thing as "Unity" gain. That is supposed to be 117 by the literature and my understanding is the high gain conversion mode starts at 120, but gain 120 is 1.000something electrons per ADU count (I would have to look up the exact figure from a .fit file) which is still marginally below unity in my book, presumably the official unity gain of 117 is lower still.
Reply With Quote
  #27  
Old 09-12-2019, 09:12 AM
Atmos's Avatar
Atmos (Colin)
Ultimate Noob

Atmos is offline
 
Join Date: Aug 2011
Location: Melbourne
Posts: 6,430
Let me put this to you.
The KAF-16803 has a well depth of 100,000e- but only a 16-bit output and typically has 13.8 bits of dynamic range.

Would you say that this old Kodak sensor has been built poorly with only a 16-bit ADC?
Or a great multitude of KAF and KAI sensors with 90k to 450k well depths not having 17 to 19-bit ADC architecture?
Reply With Quote
  #28  
Old 09-12-2019, 11:01 AM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
I can't comment on any camera I have not played with in terms of what range they can actually produce, my original question really is relevant only to the IMX294 sensor or others like it which may have a big read noise improvement somewhere in the range of gains available.

With just a bit under 2 electrons read noise (Call it 2 for the sake of it) and 65K wells (I can't see any reason to assume sensor well depth is not 65K regardless of gain) is it not reasonable to assume you could make use of 16 bits worth of conversion here?

Put it this way. From my own data it seems that at -15 degrees sensor temp and 300 second exposures, all the sources of unwanted electrons add up to under 1000e, leaving about 64ke of useable sensor range at the lowest gain that gets you the HGC mode read noise benefit. At the gain that gets the read noise benefit you can only quantize a quarter of the fuull well depth.

A big difference I can see between CCD and CMOS that means I am not sure if you can compare them in the terms done here is the use of anti blooming gates and similar techniques that mean they are not so linear near full well. The dynamic range is effectively reduced to avoid blooming artifacts so the well depth may not compare directly to CMOS wells that don't spill electrons when they are saturated. Another compromise, just a different one.

It is interesting to pinch QHY's blurb on their IMX571 based camera (QHY268C in competition with ZWO's ASI2600) It has by their page got read noise between 3.5 and 0.7e and 51ke wells.

Quote:
The QHY268C is the CMOS camera with native 16-bit A/D on-chip. The output is real 16-bits with 65536 levels. To compare with the 12bit and 14bit ADC. The 16bit ADC can get high sample resolution, system gain will close to 1e/ADU. No sample error noise and low noise.
I can't quite make head nor tail of the info ZWO have on their site now for their version, it suggest all gains on the ASI2600 will be above unity, from 0.8e/adu to about 0.05e/adu and about 100e (yes, 100 electrons!) wells at that gain! Their suggested gain of "100" where the HCG mode cuts in would seem to be about 4ADU per electron which does not seem right. Time will tell if that gets edited. But it does show nearly two stops more DR than my ASI294 at the gain where the HGC mode starts. It looks like a decimal point might be in the wrong place somewhere.
Reply With Quote
  #29  
Old 09-12-2019, 11:32 AM
Atmos's Avatar
Atmos (Colin)
Ultimate Noob

Atmos is offline
 
Join Date: Aug 2011
Location: Melbourne
Posts: 6,430
https://astronomy-imaging-camera.com...94mc-pro-color

I’m posting this link again that you clearly didn’t pay attention to.. see the FW(e-) / again graph and this shows that as gain increases full well capacity decreases. Unity gain (117) means a gain of 1e-/ADU and this LITERALLY MEANS full well depth of 16,384 e-.

In fact, those four graphs tell you answer most of your questions. Study them.
Reply With Quote
  #30  
Old 09-12-2019, 01:06 PM
Slawomir's Avatar
Slawomir (Suavi)
Amateur Photon Collector

Slawomir is offline
 
Join Date: Sep 2014
Location: Proserpine
Posts: 3,042
Interesting, that Moravian has just introduced a second line of CMOS cameras: https://www.gxccd.com/art?id=579&cat=1&lang=409
Reply With Quote
  #31  
Old 09-12-2019, 01:49 PM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
I did read that graph Colin, I have done so many times, I just interpret it differently.

As far as I am aware, the photoreceptor operates in the same way each time regardless of the gain setting and the gain is applied in an output amplifier like the block diagram below, and that makes all the difference in the world.

https://www.edmundoptics.com/content...-ucsmva-lg.gif

With the gain applied in a post pixel readout output amp, the charge capacity (Well depth in electrons) of any given pixel remains the same. Changing the gain on the output amp will change what voltage is applied to the ADC for a given number of electrons in the pixel well (Which matches the manufacturers specifying gain in terms of electrons per ADU count) Reducing gain will make the ADC less discriminating (More electrons required per ADU count change. Reduced precision in encoding) but able to quantise a greater total number of electrons in the same number of bits output. If gain is raised, it will be more discriminating (1 electron per ADU change for instance instead of 4, more precision) but the total number of electrons able to be quantised using a given number of bits drops, the sensor can continue accumulating electrons but it makes no difference to the output.

As far as I have ever been taught, and can see on any useful block diagram I have found so far, the PIXEL well depth remains the same independent of gain, but the systems effective well depth will change with gain unless the ADC bit count is high enough to quantise the whole range of voltages fromt he readout amp after gain is applied. With 64ke wells at 4 electrons per ADU, 14 bits can quantise the lot, but at 1 electron per ADU, 14 bits can only quantise a quarter of what the sensor could accumulate.

If someone can point me to info that says the number of electrons the pixel wells themselves can store changes (The number of electrons per detected photon must also change) rather than the number of electrons that the ADC can quantise changing I will be happy to be proven wrong. If not, the pixel well depth in question in that graph is the effective well depth at which point the output is saturated for the system as a whole, not the depth of the actual pixel wells on the sensor.

To put it in equivalent terms to my other expensive hobby (Race cars) The speedo in my HR1 Skyline only read to 180KMH where the new one will read to 300 if I wanted, but the original didn't stop the car from doing 220 at Phillip Island, it just couldn't display that speed.
Reply With Quote
  #32  
Old 09-12-2019, 02:57 PM
Atmos's Avatar
Atmos (Colin)
Ultimate Noob

Atmos is offline
 
Join Date: Aug 2011
Location: Melbourne
Posts: 6,430
I get what you’re saying Paul and I understand where you’re coming from there but it still comes back to inherent dynamic range.
With a well depth of 65k and 8e- read noise the sensor itself whether it be 14-bit or 16-bit ADC has about 13 stops of dynamic range. That’s it’s bottleneck. You are right that a 16-bit ADC would allow for a more accurate (near 1/1) reading of e-/ADU but there is what you could consider as 3 stops of dynamic range inaccuracy; that’s the read noise.

I’ve done some testing on a spreadsheet and the difference between 14-bit and 16-bit ADC after stacking more than a handful of exposures becomes less than one electron, it’s less than flat fielding errors, less than photon shot noise, potentially less than dark current.
In a single exposure there is likely to be more error introduced during calibration than error caused by the ADC.

Remember, if a 16-bit ADC outputs a value of 12,001 then a 14-bit will be 12,000. 12,345 becomes 12,344. 13,522 becomes 13520/13524.
Reply With Quote
  #33  
Old 09-12-2019, 03:14 PM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
It is not so much the stacking of quantisation uncertainty at low gain (Though I know people who at the deep end of things are really looking for 1e above the noise floor signals) but the fact that to get the lower read noise you have to give up the ability to measure the top three quarters of the sensors brightness range where with 16 bits you would not.

In terms of dynamic range, appearances are that where changing the gain to 120 or above on this camera restores the dynamic range of the lowest gain setting, it does that as a ratio by quartering the read noise and quartering the maximum reading, compared to lowest gain.

Anyway, I am going to watch the new 16 bit cams with interest, and see if the ZWO spec sheet makes a bit more sense in a few weeks as QHY put their blurb one way and talk about gain "approaching unity" and 1e per ADU count and the ZWO graph shows gain of above unity at the low end and extremely high gain at the other with pixel wells like a thimble. Most of all I am interested to see how they perform in the real world.
Reply With Quote
  #34  
Old 09-12-2019, 05:13 PM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
One thing coming out of this, I am going to try a different method for shooting flats given they are not dependent on gain to create a useful master flat. I am going to try longer subs at lowest gain to accumulate about half full pixel wells on the sensor itself, not the conversion.



I can't do it right now of course, the garage is so hot I can't get the sensor down anywhere near the preferred temperature.
Reply With Quote
  #35  
Old 09-12-2019, 06:37 PM
Atmos's Avatar
Atmos (Colin)
Ultimate Noob

Atmos is offline
 
Join Date: Aug 2011
Location: Melbourne
Posts: 6,430
I’d be interested as to whether they correct things well, I’ve not attempted it myself but always assumed that something could go a bit screwy with different gain settings.
Reply With Quote
  #36  
Old 09-12-2019, 06:57 PM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
Given that flats are more or less a ratio thing from divide by 1 to divide by whatever the edge illumination percentage is, it shouldn't matter. It will be interesting to compare one way or another how it corrects the field, given that is an issue with my camera.
Reply With Quote
  #37  
Old 10-12-2019, 07:52 AM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
Well that one was a failure, I will have to go back over the settings sequence generator dropped in to the flats sequence I made, about the only one I could see that could give me the ugly, lumpy result I got was if it was binned 2x2 and I missed seeing that.
Reply With Quote
  #38  
Old 14-12-2019, 07:23 AM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
Touching back on this, this is a snip from the ZWO website about the ASI2600. I am not imagining things am I? All gains are above unity according to this. It looks like the lowest gain is set up to fill more or less the whole 16 bit ADC range and everything above that has a resultant drop in effective full well.

That said, with the lower read noise than the ASI294 (About half) at lowest gain it might perform really nicely at hat gain where IMO the 0 gain images from my 294 always looked a little flat as well as noisy.

Trawling the specs this cam looks to have some interesting bits, the photos don't show it (They may be pre production images) but the drawings on the page show a ring of mounting holes on the tilt plate on a 62mm PCD, which match the QHY guider and presumably others so would allow for fixed, bolted installation to an OAG or filter wheel (If you wanted to try out NB filters on it) to be able to eliminate a threaded connection and reduce both backspacing and the opportunities to introduce tilt.
Attached Thumbnails
Click for full-size image (ASI2600.JPG)
75.6 KB8 views
Reply With Quote
  #39  
Old 14-12-2019, 08:10 AM
Merlin66's Avatar
Merlin66 (Ken)
Spectroscopy Wizard

Merlin66 is offline
 
Join Date: Oct 2005
Location: St Leonards, Vic
Posts: 7,226
Paul,
There's some good discussions going on over at SGL Re the ASI 2600.
https://stargazerslounge.com/topic/3...omment-3756023

Pity it's only available in colour.
Reply With Quote
  #40  
Old 14-12-2019, 08:10 PM
The_bluester's Avatar
The_bluester (Paul)
Registered User

The_bluester is offline
 
Join Date: Feb 2011
Location: Kilmore, Australia
Posts: 2,231
There seems to be very little by way of mono in the pipeline.


So, it looks like my read of it is not wrong. The lowest gain of both cameras is a little over unity (If you consider unity to be 1e=1ADU count) and the differences from there is more or less where in the systems range they have decided to put themselves. In one way you could say that the sensors noise profile is not better than the IMX294, at around the lowest gain of this new sensor you have about 3.4e read noise where the IMX294 has 1.7e.


I would have to say though that this one looks like a case in point for my original question. At the lowest gain (Which is pretty much equivalent to 120 gain on the 294 where the noise drops to 1.7e) the 16 bit ADC allows for 50ke before the system is saturated where the 14 bits of the 294 is toast at 16K. It has around a full stop more dynamic range than the 294 like for like gain wise. More important for me is the absolute range available as IMO, talking in stops can be misleading, the read noise floor might be near 4e at nearest to unity on the 2600 versus 1.7e for the 294, which gives the 294 a 1 stop advantage (Relatively speaking) but pretty much everything I am looking for is still going to be comfortably above the read noise and mid brightness stars should fall well within the 50ke mark and look much nicer rather than be saturated. Time will tell when people start actually getting their hands on the ASI2600 and the QHY268C and get some images out there. The fact that they seem to have very little to nothing by way of glows is also good, though I have never had an issue calibrating them out on the 294.


I would love to buy one early but finances say I am not going to be an early adopter.

Last edited by The_bluester; 14-12-2019 at 08:41 PM.
Reply With Quote
Reply

Bookmarks

Thread Tools
Rate This Thread
Rate This Thread:

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +10. The time is now 04:21 PM.

Powered by vBulletin Version 3.8.7 | Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.
Advertisement
EQ8-R
Advertisement
Astromechanics
Advertisement
Lunatico Astronomical
Advertisement
Star Adventurer
Advertisement
Testar
Advertisement
Celestron RASA
Advertisement
Bintel
Advertisement
Astronomy and Electronics Centre
Advertisement