Which CMOS camera to suit both long and short focal lengths?

[tooth_dr]

No matter how much I try to resist, I can’t and so am in the market for a cooled mono camera.

Can I image at 1500mm and 510mm with the same camera and achieve a usable sampling rate?

Is binning an option? I know CMOS does it differently.

Which model would suit me best in the ZWO CMOS range?

 

Thanks

Adam. 

[DaveS]

Are you dead set on CMOS? I had a look at the ZWO cameras and the nearest would probably be the ubiquitous 1600, but yhe pixels are a bit small for 1500mm.

How about an Atil 4120 camera? I think the pixels would be a bit more amenable to 510mm, and would bin for 1500, while still giving a usefull resolution.

[vlaiv]

I use ASI1600 at both 380mm FL and at 1600mm FL.

At 380 it gives me roughly around 2"/pixel - it is 80mm F/6 with focal reducer / field flattener x0.79 - very good for wide field (well, what I consider to be wide field, but I have trouble fitting whole Andromeda galaxy on sensor - this year hope to do a 2x1 panel to get it whole).

At 1600mm FL it is oversampling at just a tad below 0.5"/pixel (I think it's more like 0.48"/pixel or something like that). When I image at that resolution I end up binning in software (usually after calibration, prior to stacking) for resolution of close to 1"/pixel.

Binning in software is just a bit different to binning in hardware. For example, if you bin 2x2 in hardware you will end up with 4 times as large pixel (both surface/collecting area and well depth), and I believe single "dose" of read noise. With software binning it is pretty much the same, except one small difference - you end up with two "doses" of read noise - which in case of ASI1600 means 3.4e read noise instead of 1.7e (on unity gain settings). If you add 4 times 1.7e noise it goes like this: sqrt(1.7^2 + 1.7^2 + 1.7^2 + 1.7^2) = sqrt(4x 1.7^2) = 2x sqrt(1.7^2) = 2x 1.7 = 3.4e

You can effectively look at ASI1600 as having following specs (just by using software binning):

2328x1760 pixels with size of 7.6um, 14bit ADC, 16K full well (at unity gain, or 80K at the lowest gain setting), 3.4e read noise (again at unity).

Again very good specs compared to other cameras with 7.6um or similar pixel size.

Any cmos with low read noise and small pixels is suited for both long and short FL provided that you can match resolution by binning (either x2 or x3) - as signal will add normally and read noise will add in multiples of how much you bin - x2 read noise for 2x2 software bin, x3 read noise for 3x3 bin, ....

[tooth_dr]

Thanks both of you very much.

@DaveS I have seen people using lots of short subs, and I though this would work in my favour when imaging with a longish FL in case of poor guiding etc?

@vlaiv I kinda was hoping you would say what you said! On a separate aspect. If I was to image L with one 80mm and the 1600, and get RGB with a DSLR on the other 80mm, would this be a worth while exercise?

[DaveS]

I'm not sure how short you want the exposures to be, I don't go below 5min, but then I don't image below 0.86" per pixel yet.

Binning CCD increases sensitivity which doesn't happen with CMOS.

[tooth_dr]

19 minutes ago, DaveS said:

I'm not sure how short you want the exposures to be, I don't go below 5min, but then I don't image below 0.86" per pixel yet.

Binning CCD increases sensitivity which doesn't happen with CMOS.

One of the advantages of the CMOS is taking a load of 30s or 60s subs, that was my understanding!

[DaveS]

I'n not convinced by those very short exposures, especially with the huge numbers you'd need.

[kens]

Low read noise CMOS lets you bury the read noise with shorter exposures which can be an advantage if you have tracking issues. But like everything else it’s a trade off. In this case with number of subs and processing time.

[wimvb]

Generally, there's no short cut in AP. You still need about the same total integration time. CMOS have a lower number of bits (12 or 14), and you need more subs to compensate for that. You'll also need more subs to gain dynamic range. And they have a larger dark current, so even if you image from dark skies, you'll need a larger number of subs (than ccd) to get the dark noise down.

With large sensors and small pixels, cmos can generate an awefull lot of data fast. You need to match computer performance to get the best out of all that data.

As for pixel size, this can (and has been) debated a lot. In the end I wouldn't worry about it too much. While I personally would prefer to image a little oversampled rather than undersampled, both can work fine. Undersampled images can benefit from drizzle, and oversampled images from deconvolution.

[vlaiv]

1 hour ago, DaveS said:

I'm not sure how short you want the exposures to be, I don't go below 5min, but then I don't image below 0.86" per pixel yet.

Binning CCD increases sensitivity which doesn't happen with CMOS.

Binning does not increase pixel sensitivity - QE stays the same. Only thing that happens when you bin is that you add signal from adjacent pixels together. Bigger surface - more collected photons in given time - if that is what you mean by increased sensitivity - then ok, but same happens if you bin in software. Only difference between hardware and software is how much read noise there is per binned pixel vs individual pixels. If you bin 2x2 instead of having 4 read noises (each pixel having its own), with hardware binning you add one "dose" of read noise to resulting pixel (sum of 4 adjacent pixel values) while with software binning you add 2 "doses" of read noise. In this sense both software and hardware binning increase sensitivity (or better put - increase SNR at expense of resolution).

24 minutes ago, wimvb said:

CMOS have a lower number of bits (12 or 14), and you need more subs to compensate for that.

Well, number of bits is not as straight forward as comparing numbers. Take for example a few popular CCDs, all having 16 bit ADC.

Atik 460Ex - full well capacity 18K - 16 bits is wasted on that sensor - 14 bits would come very close to full well (16K), while 15 bits would be enough to record up to full well (with lots to spare). Kaf 8300 - 26K full well, again, same as previous, 15 bits is enough. Atik 4120EX - full well capacity 9000e, 13 bits would almost be enough to cover whole range (8192), while 14 bits is more than enough.

Then there is the matter of dynamic range, or to be more precise ratio of max signal to read noise. CMOS sensors have higher dynamic range than CCDs for single exposure, although they have fewer bits.

1 hour ago, tooth_dr said:

One of the advantages of the CMOS is taking a load of 30s or 60s subs, that was my understanding!

You can look at it as advantage, or disadvantage. I look at it as advantage because, while it does mean a lot more data to process and store - for me it also means less wasted data / SNR in event of lost subs, planes / satellites overhead.

There is simple comparison between CMOS and CCD in terms of sub duration.

Let's take ASI1600 and 1 minute sub, and any CCD and 16 minute sub.

Odds are that stack of 16 subs will have smaller FWHM stars with any but top tier mount. Adding 16 1 minute subs will result in same amount of light collected, while "read noise" of such sub will be x4 normal read noise, or in case of ASI1600 at unity gain - 6.8e. Maximum signal that you will record with 16 subs will be true 16bit - or 0-65536. So you get tighter stars, similar read noise to CCD cameras and higher full well capacity - stars that would saturate in 16 minute sub on CCD - will not saturate in 16 x 1 minute subs that you add together (well it will if star is strong enough, but some stars that would saturate CCD will not saturate CMOS with this approach).

If you want to compare to CCD that has lower read noise than 6.8e - like Atik 460ex with 5e read noise - just add 9 1 minute subs instead of 16 - that will give you 5.1e noise and you will have 36K full well capacity. Atik 460ex has half of that.

Thermal noise is really not an issue with ASI1600 (some other CMOS sensors might have more of it) - it is 0.0062 e/s, or in regular 1 minute exposures - 0.372e - below read noise and far below LP noise (at least where I'm imaging). If you add 16 exposures of 1 minute each you will end up with 2.44e of noise from dark current. Single 16 minute exposure from low dark current Atik 460ex will indeed give you only 0.62e dark current noise but Atik 383 (kaf8300) for example will get you ~5.34e of dark current noise for 16 minute exposure. So as you see in terms of same imaging time even dark current of ASI1600 is not the worst.

Here are sub durations that I use - 1 minute for LRGB and 4 minute for narrow band.

[tooth_dr]

Thanks guys for the info. In general I set my subs stacking just before I start work in the morning and check it occasionally when I get a gap throughout the day. Wouldn’t matter if it takes hours to process. 

 

 

 

[tooth_dr]

@vlaiv

Thanks again ??

I actually understood most of what you wrote ?

Very much open to suggestions on cameras, must be cooled and mono, but the ZWO seems pretty good, and the images testify this. Obviously a big spend and I want a significant improvement over my DSLRs. 

[coatesg]

Great explanation from @vlaiv. One thing worth adding about dark noise - it's not the actual level of dark noise, as that can be subtracted. It's the shot noise associated with it. (Proportional to the root of the measured mean, and inversely proportional to the number of samples).

In the case given, comparing like for like, of course you still get less shot noise in the dark signal of the 460ex, but it's the overall S-N ratio that needs to be considered, including read noise (which for the Atik is at least double, if not 5x the 1600, depending on gain).

Back to the OP, I have a 163M and plan on using it for both camera lens work and on the big 14" newt which has FL=1583mm. I think 2x binning is sensible on the newt, unless I decide to try lucky imaging on there with v short exposures to try to beat the seeing.

[ollypenrice]

The stacking time varies enormously between software packages. The fastest I know is also the one I use - AstroArt - which I think might be of interest to CMOS users because of its speed. I also think it does a good job and is admirably transparent and adjustable in terms of what it is going to do during the process. A free 'no save' version is available for trial.

Olly

[Craney]

19 hours ago, tooth_dr said:

No matter how much I try to resist, I can’t and so am in the market for a cooled mono camera.

So then, Adam..... which one of us will break first ??

[tooth_dr]

2 minutes ago, Craney said:

So then, Adam..... which one of us will break first ??

I literally just received an email, with a remittance for £1500, about an hour ago, for a training programme I took part in last year.  Is this a coincidence?

[Craney]

I await the decision with great interest.

(my green channel colour balance has just saturated...)

[Adam J]

20 hours ago, vlaiv said:

I use ASI1600 at both 380mm FL and at 1600mm FL.

At 380 it gives me roughly around 2"/pixel - it is 80mm F/6 with focal reducer / field flattener x0.79 - very good for wide field (well, what I consider to be wide field, but I have trouble fitting whole Andromeda galaxy on sensor - this year hope to do a 2x1 panel to get it whole).

At 1600mm FL it is oversampling at just a tad below 0.5"/pixel (I think it's more like 0.48"/pixel or something like that). When I image at that resolution I end up binning in software (usually after calibration, prior to stacking) for resolution of close to 1"/pixel.

Binning in software is just a bit different to binning in hardware. For example, if you bin 2x2 in hardware you will end up with 4 times as large pixel (both surface/collecting area and well depth), and I believe single "dose" of read noise. With software binning it is pretty much the same, except one small difference - you end up with two "doses" of read noise - which in case of ASI1600 means 3.4e read noise instead of 1.7e (on unity gain settings). If you add 4 times 1.7e noise it goes like this: sqrt(1.7^2 + 1.7^2 + 1.7^2 + 1.7^2) = sqrt(4x 1.7^2) = 2x sqrt(1.7^2) = 2x 1.7 = 3.4e

You can effectively look at ASI1600 as having following specs (just by using software binning):

2328x1760 pixels with size of 7.6um, 14bit ADC, 16K full well (at unity gain, or 80K at the lowest gain setting), 3.4e read noise (again at unity).

Again very good specs compared to other cameras with 7.6um or similar pixel size.

Any cmos with low read noise and small pixels is suited for both long and short FL provided that you can match resolution by binning (either x2 or x3) - as signal will add normally and read noise will add in multiples of how much you bin - x2 read noise for 2x2 software bin, x3 read noise for 3x3 bin, ....

The thing with binning the ASI1600mm at 1600mm FL  is that you could get a similar result by using it at 800mm without binning it and benefit from a much lighter (potentially faster) easier to Handel scope. So something like a F5 / F6, 6 inch Newtonian. You also would not suffer from the binning losses due to the additional read noise per (virtual) pixel.

Am I missing something? As far as I am aware a 6 inch aperture is easily able to resolve down to ~1arcsec / pixel...resolution in the final image is likely limited by seeing and so in fact a faster shorter scope might even help because you could image something like a galaxy by using the v.short exposure method. 

 

[vlaiv]

1 minute ago, Adam J said:

The thing with binning the ASI1600mm at 1600mm FL  is that you could get a similar result by using it at 800mm without binning it and benefit from a much lighter (potentially faster) easier to Handel scope. So something like a F5 / F6, 6 inch Newtonian. You also would not suffer from the binning losses due to the additional read noise per (virtual) pixel.

Am I missing something?

 

You are quite correct, but there are a couple of things that are different to my current setup if you go that route:

- less aperture. 6" scope will collect less light than 8" scope at 1"/pixel resolution. Depending on target brightness and exposure length this translates in lower SNR for a given imaging time (even if we account for double the read noise when binning) - it will how ever have potential of wider field.

- I use RC8" without any corrective optics, and stars look good up to edge of sensor (ASI1600). Fast(ish) Newtonian in F/6 or F/5 class will not be able to do that, so coma corrector is in order. From what I've seen, CCs sometimes have a tendency to make stars softer - thus loosing resolution (not all but some do. there are however more expensive versions that would do excellent job even on 8" F/4 - F/5 scope with focal reduction and still offer resolution of 1"/pixel and good looking stars, but cost of CC is almost as much as my scope alone). In case of optically very good 8" F/2.8 astrograph, even smaller pixels of cameras like ASI183 would provide high native resolution (0.88"/pixel) again at 8" collecting. But such scopes are pretty expensive, and I would consider something like that as fast EAA rig alongside imaging.

- OTA length will certainly be more even with F/4 scope compared to my OTA - more strain on mount and guiding. Other "perks" like enough back focus for accessories, 99% dielectric coatings somewhat compensate for large CO (but imaging newtonians have that also to get good field illumination)

- Newtonians don't handle stray light as good as RC8" (fully baffled) so tweaks are needed there (blackening behind focuser, extending ota with dew shield, ...) .

Initially when I was looking at imaging OTA to put on HEQ5, fast newtonian was sensible choice, and F/8 of RC was considered "Slow", but after doing the math, I decided that it is not only viable option, but I consider it better (for my needs) given above points.

[tooth_dr]

So just to clarify before I buy this camera —>

I have both a 10” and a 12” newt, both a smidge under F/5, and a CC, id like to image galaxies etc.

I also have an ED80 600mm with 0.85x FF, for nebula etc. 

Good choice with the ASI1600MM?

[vlaiv]

I would say yes if your mount is up to it (and I guess it is).

12" newt will be very fast (lots of aperture) - just use software bin 2x2 for resolution of 1.05"/pixel.

10" newt will be good for those larger galaxies that don't fit in FOV of 12" - like M81, M33 or M101 - it will give you around ~1.3"/pixel when binned 2x2 (again good resolution for larger galaxies, or in general when the seeing is poor).

ED80 will cover everything else with resolution of 1.54"/pixel - but you will not be able to image M31 in one go - it will not fit on sensor, at least two panels will be needed (as well as some larger nebulae in NB - those also may need multiple panels).

[tooth_dr]

18 minutes ago, vlaiv said:

I would say yes if your mount is up to it (and I guess it is).

12" newt will be very fast (lots of aperture) - just use software bin 2x2 for resolution of 1.05"/pixel.

10" newt will be good for those larger galaxies that don't fit in FOV of 12" - like M81, M33 or M101 - it will give you around ~1.3"/pixel when binned 2x2 (again good resolution for larger galaxies, or in general when the seeing is poor).

ED80 will cover everything else with resolution of 1.54"/pixel - but you will not be able to image M31 in one go - it will not fit on sensor, at least two panels will be needed (as well as some larger nebulae in NB - those also may need multiple panels).

 

Thanks for this answer! I will owe you a ?

[Adam J]

2 hours ago, tooth_dr said:

So just to clarify before I buy this camera —>

I have both a 10” and a 12” newt, both a smidge under F/5, and a CC, id like to image galaxies etc.

I also have an ED80 600mm with 0.85x FF, for nebula etc. 

Good choice with the ASI1600MM?

It will work well.

[coatesg]

3 hours ago, Adam J said:

The thing with binning the ASI1600mm at 1600mm FL  is that you could get a similar result by using it at 800mm without binning it and benefit from a much lighter (potentially faster) easier to Handel scope. So something like a F5 / F6, 6 inch Newtonian. You also would not suffer from the binning losses due to the additional read noise per (virtual) pixel.

Am I missing something?

 

For a given camera, I think the big Newt still wins out (mechanical/optical considerations aside). For a given focal length, the quicker scope has higher SN ratio and that would be the larger scope (aperture wins here).

NxN Binning with the cmos camera, by theory should give ~ an N^1/2 increase in SNR again unbinned, but comparing different scopes binned Vs unbinned probably won't hold in practise.

It would be much less demanding on the mount using a shorter scope, agreed. Not sure I understand what you mean by binning losses though, unless talking about total number of pixels?

None of this means either situation won't work of course - there are plenty of examples out there to show either situation gives great results!

[tooth_dr]

Thanks everyone for your input!

Final thought. We are on the 4th gen now of this model. Should I wait for the next one? Lol