ZWO Bit depth question.

[ollypenrice]

I've read a couple of owners of the ZWO ASI 1600MM Cool comment to the effect that multiple stacking restores the bit depth. I'm not doubting them but don't know enough about the electronics to understand this. Anybody care to help me out? The images coming in are very impressive and the price is more than reasonable.

But retailers, please,

give us the chip size in mm for each side!!!!!!!

So you get the bit depth back by multiple stacking, but how does this work?

Olly

 

 

 

[Martin Meredith]

This article by Craig Stark explains how it is possible (thanks to noise!).

Martin

[ollypenrice]

Just now, Martin Meredith said:

This article by Craig Stark explains how it is possible (thanks to noise!).

Martin

Thank you, kind Sir!

Olly

[MrsGnomus]

1 hour ago, ollypenrice said:

But retailers, please,

give us the chip size in mm for each side!!!!!!!

Sunday morning Pythagoras theorem isn't necessarily my strong point but if the diagonal is 21.9 mm and using the information that the chip is 4656 x 3520 pixels it should follow that the long side if the chip is 1.323x the length of the short side. I worked out that this would make the chip 13.2 by 17.46 mm. 

@gnomus suggested that it would be simpler to just multiply the pixel size by the number of pixels quoted for each dimension - this gives a slightly different answer of 13.376 by 17.6928 - this would give a predicted diagonal of 22.18 which isn't the quoted size!

Not sure which of us is correct or if the tiny difference in results even matters....

 

[ollypenrice]

Just now, MrsGnomus said:

Sunday morning Pythagoras theorem isn't necessarily my strong point but if the diagonal is 21.9 mm and using the information that the chip is 4656 x 3520 pixels it should follow that the long side if the chip is 1.323x the length of the short side. I worked out that this would make the chip 13.2 by 17.46 mm. 

@gnomus suggested that it would be simpler to just multiply the pixel size by the number of pixels quoted for each dimension - this gives a slightly different answer of 13.376 by 17.6928 - this would give a predicted diagonal of 22.18 which isn't the quoted size!

Not sure which of us is correct or if the tiny difference in results even matters....

 

Très gentil, madame! I did much the same sum. The discrepency may be to do with active pixel area, perhpas. But why don't they just give the blessèd information since we so obviously need it!

Olly

[MrsGnomus]

4 minutes ago, ollypenrice said:

But why don't they just give the blessèd information since we so obviously need it!

I agree it would be much simpler for all concerned for them to give us the information but perhaps they are getting some perverse pleasure in forcing us to revisit maths that we learned decades ago!

[wimvb]

Re bit depth, as I understand it, the source of the noise doesn't matter. It doesn't need to be the camera, or does it? I don't want to start shopping for a 'moderately' noisy camera.

Re sensor specs; why is it that read noise is always specified, but rarely dark current/noise? They could specify it for a certain temperature. I would like to compare cmos dark current/noise against ccd.

Or am I missing something again?

@MrsGnomus, some of us still teach it 

[Martin Meredith]

If it is really 'noise' and not fixed pattern 'noise', I don't think it matters where it is generated, so long as it is prior to the analogue-to-digital conversion.

There's an interesting paragraph and figure (10) in the document I reference which shows how much noise is ideal too.

I see the situation as somewhat analogous to the way spatial noise (dithering) can be used to increase spatial resolution.

Martin

[Owmuchonomy]

When I was researching these cameras a while ago the ZWO website spec said the ASI 1600 mono cooled chip is 17.6mm x 13.3mm (image capture area).  

[MrsGnomus]

34 minutes ago, wimvb said:

@MrsGnomus, some of us still teach it 

Oh no! Does that mean that my "homework" will end up being marked?  Perhaps I should have waited for son number 3 to get home - he is the aspiring household mathematician.

[wimvb]

25 minutes ago, MrsGnomus said:

Oh no! Does that mean that my "homework" will end up being marked?  Perhaps I should have waited for son number 3 to get home - he is the aspiring household mathematician.

Maybe he can answer the following question:

Stark uses average as the stacking method, and that works fine. How would his method translate to 'median' stacking? Afaik, median stacking uses one of the actual pixel values to compose the final image. How can this lead to an increase in bit depth? Common sense would indicate that median follows mean (average), but here we are referring to discrete, integer values, not fractions as with averaging.

@Martin Meredith, you are correct of course, noise can only mean the randomness in the numbers. But if a target registers on different pixels, the alignment process will see to it that fixed pattern 'noise' will be interpreted as randomness. This is in fact like the dithering (and drizzle) you also refer to.

 

[StuartJPP]

I am sure that there is (inactive) spacing between pixels so that probably explains some discrepancies. If the active pixel is specified exactly (e.g 3.75um) and then you multiply that by the resolution it will not match the sensor size because of not taking the spacing into account. Additionally there are occasionally rows/columns of pixels that are not presented to the user and sometimes they are but only in a raw format. They can be used to estimate dark currents.

The datasheet for the Panasonic MN34230 sensor has the following items. I could only find the colour version but it is identical to the mono barring the Bayer matrix...

 

With regards to increasing the bit depth, it all relies on shot noise from what I remember and stacking multiple sub frames will increase the bit depth to some degree which involves maths way beyond my capabilities.

[wimvb]

That's a quite a detailed specsheet. Now you'd just have to figure out the difference between effective and active pixels. And are the rest inactive?

As for noise: shot noise is the common name for any noise that has to do with quantization effects, whether it is photons hitting the sensor, or electrons passing a barrier/moving out of the well. It stems from the fact that a certain number of events is likely to happen, but you can't know exactly when or how many will happen. Shot noise comes from any light source illuminating the sensor, as well as the electronics on the sensor.

Since Stark in his article doesn't specify the source of the noise, any noise in a calibrated and aligned image should do.

If I understand his argument correctly, if the combined noise in an image is large enough, the bit depth of a camera becomes less important (Starks figure 10). Noise of 0.5 ADU in an 8 bit camera seems little, but if it translates to 128 ADU in a 16 bit camera, it suddenly becomes a lot. It would be interesting to see the comparison of 12-, 14-, and 16-bit cameras.

[StuartJPP]

Wim...I told you it is beyond my capabilites

I realise that it pertains to more than just shot noise having now read and partially understood what Craig is on about. However from my limited understanding that is the most important noise source and where it all starts since if you didn't have a chance of collecting it in the first place all other sources of noise in the system would certainly not be signal, it would just be noise. So without shot noise you would end up with a cleaner image (greater signal to noise) but you wouldn't be able to increase the bit depth beyond 12 bits for example.

Of course I may be speaking out of my rear end again...

 

[Martin Meredith]

Shot noise isn't a necessary condition for bit depth to increase. Stark uses Gaussian noise in his simulations.

To understand the role of noise, we only need to consider a single pixel. Suppose the true value of that pixel is 21.6. After quantisation, and without noise, it is always quantised as 22. We can stack as many copies as we like but we still get that value. But if there is some noise (e.g., from seeing), some of the time the pixel value will be nearer to 21 and some time nearer to 22. In fact, it will be nearer to 22 more times than it will be nearer to 21. How many more times? Well, since the true value is 21.6 we can expect on average 6 times out of 10 to get a value of 22. By stacking, we therefore arrive at a better estimate of its true value. In the same way the true value might be 21.67, and after a stack of 100 we arrive at a good estimate of that value. And so on. By stacking more we increase bit depth more.

Martin

[sharkmelley]

The other information I would be interested in is the QE (Quantum Efficiency).

It's odd that a camera can be sold to the astro community without such information.

Mark

[wxsatuser]

5 hours ago, StuartJPP said:

I am sure that there is (inactive) spacing between pixels so that probably explains some discrepancies. If the active pixel is specified exactly (e.g 3.75um) and then you multiply that by the resolution it will not match the sensor size because of not taking the spacing into account. Additionally there are occasionally rows/columns of pixels that are not presented to the user and sometimes they are but only in a raw format. They can be used to estimate dark currents.

The datasheet for the Panasonic MN34230 sensor has the following items. I could only find the colour version but it is identical to the mono barring the Bayer matrix...

 

With regards to increasing the bit depth, it all relies on shot noise from what I remember and stacking multiple sub frames will increase the bit depth to some degree which involves maths way beyond my capabilities.

It's a four thirds found in the likes of these cameras E-M1, GH4, GX7, GM1

[wxsatuser]

42 minutes ago, sharkmelley said:

The other information I would be interested in is the QE (Quantum Efficiency).

It's odd that a camera can be sold to the astro community without such information.

Mark

The QE for these cameras, E-M1, GH4, GX7, GM1  comes out between 43 to 56%, would assume the mono version could be better.

[ollypenrice]

As ever with all these numbers, what does it really do on the sky? I'm seriously thinking of trying one of these. Since most of our imaging is done at a contemptible 50% QE and an even more contemptible 3.5 arcsecs per pixel I'm not always intimidated by bad numbers.

Olly

[harry page]

Hi

These chips are harder to cool than a ccd and the dark current is higher than a ccd , also the amp glow is still a problem in longer exsposers

They can be read very quickley ( many times a second ) and are therefore versitile .

I had a 12 bit camera many years ago and found it a struggle , I think the highest a cmos will go is 14 bit , so get uded to not having 16 bit cameras

Personally I will wait till further improvements are made with cmos chip cameras , which I know are coming as ccd cameras still out perform cmos cameras

Regards

Harry

[wimvb]

30 minutes ago, harry page said:

Hi

These chips are harder to cool than a ccd and the dark current is higher than a ccd , also the amp glow is still a problem in longer exsposers

They can be read very quickley ( many times a second ) and are therefore versitile .

I had a 12 bit camera many years ago and found it a struggle , I think the highest a cmos will go is 14 bit , so get uded to not having 16 bit cameras

Personally I will wait till further improvements are made with cmos chip cameras , which I know are coming as ccd cameras still out perform cmos cameras

Regards

Harry

Therefore my earlier question as to why dark current is rarely specified for these cameras. On the other hand, I've seen some good results with exposure times shorter than a minute. Both dark current and amp glow are less of an issue then.

ZWO have addressed amp glow (according to the user forum). But there is still some in the newest model. It seems that this can be calibrated out.

Following posts in this forum, zwo forum and cloudy nights, issues with software and drivers are a recurring theme.

If you can do shorter exposures and more of them, then the above discussion about bit depth becomes very relevant. (which is what started this thread)

[Andyb90]

1 hour ago, ollypenrice said:

I'm seriously thinking of trying one of these.

I'd say go for it, although I'm slightly biased . As you say the price is very reasonable and I'm sure you'd have no problem selling it on if you found the camera wasn't for you.

Andy.

[StuartJPP]

3 hours ago, sharkmelley said:

The other information I would be interested in is the QE (Quantum Efficiency).
It's odd that a camera can be sold to the astro community without such information.
Mark

There's some detail in the user manual:
http://astronomy-imaging-camera.com/manuals/ASI1600 Manual.pdf

 

I love the graph almost purporting to be 100% efficient at 530nm until you realise that the peak QE is 60%...

[sharkmelley]

8 hours ago, harry page said:

These chips are harder to cool than a ccd and the dark current is higher than a ccd , also the amp glow is still a problem in longer exsposers

 

What makes them harder to cool than a CCD?

Is the ASI 1600 dark current really higher than CCD?  I haven't seen figures for the ASI 1600 sensor in particular and I wholeheartedly agree that for an astro-camera they really ought to be published.  However the popular KAF-8300 CCD sensor has a dark current of around 1.5e/sec at 20C whilst Sony Exmor CMOS has a dark current of around 0.15e/sec at 20C.  I don't know how the ASI 1600 sensor compares with Sony Exmor.

Mark

[ollypenrice]

10 hours ago, harry page said:

Hi

These chips are harder to cool than a ccd and the dark current is higher than a ccd , also the amp glow is still a problem in longer exsposers

They can be read very quickley ( many times a second ) and are therefore versitile .

I had a 12 bit camera many years ago and found it a struggle , I think the highest a cmos will go is 14 bit , so get uded to not having 16 bit cameras

Personally I will wait till further improvements are made with cmos chip cameras , which I know are coming as ccd cameras still out perform cmos cameras

Regards

Harry

Yes, there's a lot to be said for waiting because they're going to improve for sure.

Olly