•

# ASI1600MM Camera Performance

This topic has been archived. This means that you cannot reply to this topic.
79 replies to this topic

### #51 mikefulb

mikefulb

Surveyor 1

• Posts: 1,884
• Joined: 17 Apr 2006

Posted 17 June 2016 - 05:04 PM

Given the low read noise (relative to the 8300) and similar QE it actually isn't a disadvantage.  If you collected N hours of 300s exposures with the 1600 and N hours of, say, 1800s exposures with the 8300 in a dark site I believe statistically both would give you close to photon statistic limited results.  It doesn't take much exposure to overcome 2 e- of read noise for the 1600 compared to 8 e- (or more) with 8300 based cameras.

Edited by mikefulb, 17 June 2016 - 05:04 PM.

### #52 Jon Rista

Jon Rista

ISS

• Posts: 24,854
• Joined: 10 Jan 2014

Posted 17 June 2016 - 05:10 PM

Well, the terms noise and signal are often used interchangably. For one man a signal may be a signal, and for another it may be a noise.

In the case of astrophotography. Dark current is a signal, but it is also a noise, and it contains non-random pattern noise. The bias signal is a signal, but it's also a noise, and contains non-random pattern noise as well. Glows are a signal, but they are also a noise. Skyfog is a signal, but it's also a noise. Read noise is actually just a noise. There is a signal that produces it, however that signal never enters the image, only it's noise does (and that is the reason why the term must be squared before being added to other noise terms.)

Dark current has it's own random noise. A shot noise. It's the accumulation of electrons (discrete events) over time. So that shot noise follows a Poisson distribution. It's noise is the square root of the signal. So if you have 100e- dark current, then you have SQRT(100) or 10e- dark current noise. We can subtract the extra signal. That 100e- offset. However, it's noise, random variations from the expected additional 100e- offset, will combine with other noises. It's SQRT(random+random+random+...)...you can't know what it is, so you can't directly remove the noise.

Thermal glows are similar to dark current, and actually related to it. A glow is actually the result of non-uniform dark current rates across a sensor. Since dark current rate is relative to temperature, if something produces local heating in certain areas of the sensor, then the dark current will be higher in those areas. There can be other causes of glow as well. Differences in voltage response, due to the manufacturing and materials of the sensor. Those are usually fixed bias glows. Change in voltage from the voltage source of a column of pixels to the opposite end of that column can create a voltage gradient, resulting in a visible gradient in the bias signal. Such a glow can change, especially in a sensor that self calibrates the column (or row or even pixel) voltages when it's powered up. Unfiltered sources of infrared radiation can result in non-thermal glows (I think some other ASI cameras might actually suffer from this, given the ray-like nature of some parts of their glows.) Like dark current, these are all signals, the accumulation of electrons (discrete events) over time. So, they follow a Poisson distribution. The noise is the square root of the signal. So if you have localized pockets of 150-300e- glows, then you have SQRT(150) to SQRT(300) noise, but localized to those areas. We can subtract these localized extra signals. The 150-300e- offsets. However, their noise, random variations from the expected additional 150-300e- offset, will combine with other noises. It's SQRT(random+random+random+...)...you can't know what it is, so you can't directly remove the noises.

Skyfog is another signal that enters through the aperture. It can be fairly uniform across the field, or it may be non-uniform depending on where light pollution sources are and how large the field of view is. For a given exposure the flux is usually consistent for each pixel, but a pixel in one corner may acquire fewer photons than in another. So skyfog tends to present as a gradient. It also has photon shot noise, like dark current, like your object signal. This signal could be tens to thousands of electrons in size, depending on how much light pollution you have. As this is the accumulation of electrons in response to incident photons on the pixels (discrete events) over time, it follows a Poisson distribution. It's noise is the square root of the signal. We can model the gradient and subtract it from the image. Like other forms of noise, it's SQRT(SkyFog), it's random, so it will combine with other noises. It's SQRT(random+random+random+...)...you can't know what it is, so you can't directly remove the noise.

These are all signals, all of these signals we remove from the image. Either with calibration, as in the case of bais and dark current and glows. We subtract master bias and master dark frames to eliminate the offsets these extra signals add to our pixels. We model gradients ("background sky") in our images, and subtract those gradients to remove the offset light pollution adds to our pixels. So, we can call these signals "noise". They are unwanted artifacts. They interfere with and interrupt our desired signal. So noise is an apt description. But there are lower level mechanics in play, and when it comes to understanding what's actually going on, it helps to understand that even though these things are noise in relation to the signal we are actually interested in preserving, each of them in turn is a signal WITH a noise component, and that while these signals can be removed, their noises cannot be.

The only thing we can do with true noise...the random variations in pixel values...is to average them out.

### #53 terry59

terry59

Voyager 1

• Posts: 10,401
• Joined: 18 Jul 2011

Posted 17 June 2016 - 05:23 PM

Jon - say we're going after something really faint and small and 300s isn't adequate - are the glows restricted to the periphery of the chip sufficiently that if we cropped down to the central 50% of the sensor we could maintain a steady increase in S/N for that area?  I'm thinking interacting galaxies or OII/SII data for the outer shells of M57.

If you crop down to the central 50%, you wouldn't have the severe glows.

50%!!!! Ummm....really???

Edited by terry59, 17 June 2016 - 05:26 PM.

### #54 Jon Rista

Jon Rista

ISS

• Posts: 24,854
• Joined: 10 Jan 2014

Posted 17 June 2016 - 05:25 PM

Jon - say we're going after something really faint and small and 300s isn't adequate - are the glows restricted to the periphery of the chip sufficiently that if we cropped down to the central 50% of the sensor we could maintain a steady increase in S/N for that area?  I'm thinking interacting galaxies or OII/SII data for the outer shells of M57.

If you crop down to the central 50%, you wouldn't have the severe glows.

50%!!!! Ummm....really???

Heh. You could also just crop out the outer 10% (and probably less than that), which would eliminate the glows. You would still have a slight bit of gradient, but even CCDs have those, including the new KAF-16200. I was responding in context...if you cropped down to the middle 50% when imaging M57, you would certainly eliminate the glows.

Edited by Jon Rista, 17 June 2016 - 05:25 PM.

### #55 syscore

syscore

Aurora

• Posts: 4,711
• Joined: 07 Sep 2015

Posted 17 June 2016 - 06:39 PM

Jon, I think you are LP limited. The noise in your LP is greater than the (very faint) signal you are trying to measure, and thus, you cannot (ever) measure it.

Edited by syscore, 17 June 2016 - 06:39 PM.

### #56 syscore

syscore

Aurora

• Posts: 4,711
• Joined: 07 Sep 2015

Posted 17 June 2016 - 07:16 PM

"OK consider unwanted signals as a form of noise isn't a real bad idea though, just a type of removable noise."

When a pixel records how much light has fallen on it, the value it stores is not exact. One time it might read 100, and another 95 and another 105. This isn't the fault of the sensor, it is due to the physics of measuring a signal. There is always an error term in the measurement.

Pixels also (unfortunately) measure heat, an "unwanted" signal, and those measurements also have an inherent error in them as well.

It is these errors in the measurements we call noise. If the measurements were exact then we could subtract all of the unwanted signals, easily, by simply sliding the left slider to the right, and we would be left exactly with the signal we were after. But none of the measurements are exact and the errors compound, and when we slide the slider to remove the unwanted signals, we don't see black, we see that noise.

The only way around this is to increase the total integration time because (fortunately) the errors in these signals, wanted or not, increase by SQRT(TIME) while the signals themselves increase by TIME. Thus, these errors, as a percentage, become smaller.

However, if some of the unwanted signals, like LP or amp glow, are relatively large to begin with, then the errors in them are also relatively large. The larger the unwanted signals then the larger the error associated with them and when we subtract them we are left with that error and that error causes the dimmer parts of our images to be noisy or even hidden.

Finally, when we read the measurement from the pixels there is an error associated with that. A read error, which we call read noise. Fortunately, that does not change with TIME, but it does happen every time we read the pixel, and when we use shorter subs, we read the pixel more often and the total read noise becomes greater.

Short answer. The larger the unwanted signals, then the larger the errors associated with their values and the less our ability to remove them from the wanted signal. We counter this by making the unwanted signals smaller when we can, by cooling the sensor or imaging in darker skies, and by imaging longer to make the errors smaller in comparison. And finally, we control the read error by taking longer subs and reading the sensor less often.

Edited by syscore, 18 June 2016 - 05:52 AM.

### #57 syscore

syscore

Aurora

• Posts: 4,711
• Joined: 07 Sep 2015

Posted 17 June 2016 - 07:22 PM

Do you mean for my AF?

AF = auto focus?

No, I mean the reason you are struggling with the dim detail in M57 is that it is very dim and the noise implicit in your LP at your location is hiding it. Your result looks just like this one...

http://www.cloudynig...ject/?p=7277232

And that is simply due to noise. Since we know that the camera isn't adding the noise that is hiding the detail, then it must be your LP. I realize that you are using an NB filter, but that still lets in LP, and even though it is a small amount of LP, what you are trying to image (the dim details in M57) is very dim.

I don't think focus is enough to explain your results.

jmo

Edited by syscore, 17 June 2016 - 07:24 PM.

### #58 syscore

syscore

Aurora

• Posts: 4,711
• Joined: 07 Sep 2015

Posted 17 June 2016 - 09:10 PM

"You think I am getting skyfog limited subs with a 3nm Ha filter in a mere 300 seconds? Seriously?"

When you image, whatever noise there is, settles at the floor. If the magnitude of the dimmest detail is not sufficiently above that noise, then it will either be a haze or not visible at all.

You are thinking in terms of the limits of the whole image. I am only talking about the dimmest details in that image. Tolga was at f/6, I at f/7 and you at f/8. We exposed for 1 hour and 15 minutes, while you exposed for 3 hours. Also, you said that longer exposures didn't help. You have something blocking the dimmest detail.

I am just saying that the there is noise at the bottom that is obscuring the dim detail and the dim stars. Hiding it even.

I am assuming it is coming from the sky, since it isn't coming from the camera.

I am not talking skyfog limit, and even if you exposed to the skyfog limit, your total noise from all sources would be some number, say, 200 e-, then any detail close to that will be hazed, and less than that will be invisible.

### #59 FirstC8

FirstC8

Apollo

• Posts: 1,118
• Joined: 10 Mar 2016

Posted 17 June 2016 - 09:31 PM

Syscore, thanks for the explanation on what is signal and what is noise.

One question, why is that a stronger signal will result in larger read error (read noise)?

Use M57 as example, for a certain exposure time, although the signal from the core is much, much stronger than that from the halos, you are not suggesting read noise is also much higher at the core than at the halos?

Edited by FirstC8, 17 June 2016 - 09:35 PM.

### #60 Goofi

Goofi

Cosmos

• Posts: 8,137
• Joined: 03 May 2013

Posted 17 June 2016 - 09:33 PM

Jon (and everyone else), I think syscore makes a good point. I'm struggling with this on an image I'm working on right now.

I've done a screen-grab to show what's happening. The left half is my Ha channel, stretched. The right half is the same image, inverted. In the inverted image you can see a lot more structure (top-center, for instance) ... but it's up against the noise floor and pulling it out is challenging. I need more subs (to crush noise) or to go deeper.

### #61 syscore

syscore

Aurora

• Posts: 4,711
• Joined: 07 Sep 2015

Posted 17 June 2016 - 09:43 PM

"One question, why is that a stronger signal will result in larger read error (read noise)?"

If a signal is 100, then the error in that signal (dictated by physics) is SQRT(100) = 10.

If a signal is 10000, then the error in that signal is SQRT(10000) = 100.

The error isn't actually SQRT(S), the error is a distribution, but the standard deviation of that distribution is SQRT(S) which is a good way to measure it.

In any event, the error varies as the square root of the signal.

The good thing is that the SNR, which is what you see as CONTRAST, is S / SQRT(S), and since the error only goes up by the SQRT(S) then the longer you integrate the higher the SNR.

If that were not the case and the error went up linearly with the signal, well, this would probably be a pottery forum, because astrophotography would be impossible.

Edited by syscore, 17 June 2016 - 09:45 PM.

### #62 syscore

syscore

Aurora

• Posts: 4,711
• Joined: 07 Sep 2015

Posted 17 June 2016 - 09:51 PM

Use M57 as example, for a certain exposure time, although the signal from the core is much, much stronger than that from the halos, you are not suggesting read noise is also much higher at the core than at the halos?

Read noise is a different term. It only occurs when you read the sensor. It is the same all over the image. If it is 2 e- for example, then every pixel read has an error of +/- 1 on average. That isn't the exact number, but close enough.

And every time you read again (each sub exposure), you add another read error into the mix. The addition though is SQRT(2^2 + 2^2) = SQRT(8) = 2.8

The next sub = SQRT(2^2 + 2^2 + 2^2) = SQRT(12) = 3.5

And so on.

To limit total read noise you limit how many times you read, and to limit how many times you read, you make your sub exposures longer.

But read noise does not depend on the signal. It is an error associated with the electronics and how it reads the values from the sensor. It is the same whether the sub exposure was 1 second or 1 hour.

### #63 syscore

syscore

Aurora

• Posts: 4,711
• Joined: 07 Sep 2015

Posted 17 June 2016 - 09:57 PM

And the only point I am making is that we are looking at the dimmest details here. We are right next to the noise. There is a limit as to how deep any one of us can get, and the sky plays a role in that. Anything that adds noise plays a role in that.

### #64 FirstC8

FirstC8

Apollo

• Posts: 1,118
• Joined: 10 Mar 2016

Posted 17 June 2016 - 10:03 PM

"One question, why is that a stronger signal will result in larger read error (read noise)?"

If a signal is 100, then the error in that signal (dictated by physics) is SQRT(100) = 10.

If a signal is 10000, then the error in that signal is SQRT(10000) = 100.

The error isn't actually SQRT(S), the error is a distribution, but the standard deviation of that distribution is SQRT(S) which is a good way to measure it.

In any event, the error varies as the square root of the signal.

The good thing is that the SNR, which is what you see as CONTRAST, is S / SQRT(S), and since the error only goes up by the SQRT(S) then the longer you integrate the higher the SNR.

If that were not the case and the error went up linearly with the signal, well, this would probably be a pottery forum, because astrophotography would be impossible.

The rule you used is a function of time though, as you continue to accumulate signal, the noise also accumulate but at a slower rate.

But my question is about your statement that if a signal itself is stronger (not time based), so is the read error (noise). Maybe you meant only unwanted signal and the corresponding read error (noise)?

I hate to think a wanted strong signal also results in higher noise.

Edited by FirstC8, 17 June 2016 - 10:06 PM.

### #65 Goofi

Goofi

Cosmos

• Posts: 8,137
• Joined: 03 May 2013

Posted 17 June 2016 - 10:09 PM

Note from the moderator ....

There were some posts posted to this thread on accident. I was asked by the originator to move them to the correct thread and I did, along with direct replies. You can find the moved posts here. No posts were deleted, just moved.

### #66 syscore

syscore

Aurora

• Posts: 4,711
• Joined: 07 Sep 2015

Posted 17 June 2016 - 10:12 PM

"But my question is about your statement that if a signal itself is stronger (not time based), so is the read error (noise). Maybe you meant only unwanted signal and the corresponding read error (noise)?"

For any signal, the noise is the square root of the signal, so if the signal is larger then the noise will be larger.

Which means, if I have a detail next to a bright star, that is dimmer than the noise caused by that star, then I can't see it. If it were not for that, then we could adjust our sliders and see practically anything. It is noise that limits our sliders. And signal noise is unfortunately a fact of nature.

Read noise is something else, and small, and constant, and per exposure.

Edited by syscore, 17 June 2016 - 10:14 PM.

### #67 Jon Rista

Jon Rista

ISS

• Posts: 24,854
• Joined: 10 Jan 2014

Posted 17 June 2016 - 10:21 PM

FirstC8, just to make sure it's clear. These rules apply to Poisson signals. Discrete events over time (photons striking pixels, electrons accumulating in pixels due to dark current, etc.) For such a signal, the SNR is always Signal/SQRT(Signal). If you acquired 100 photons, your noise would be SQRT(100), as would your standard deviation. If you acquired 1000 photons, your noise would be SQRT(1000). The signal increased by a factor of 10, however the noise increased by a factor of SQRT(1000)/SQRT(100), or only by a factor of 3.16x. So yes, signal grows faster.

In the case of astrophotography, however, SNR for our images is not just Signal/SQRT(Signal). There are other noise terms that add noise, but do not add to signal.

So if you have 100e- signal, and 180e- glow, 5e- dark current, 50e- skyfog, and 1.56e- read noise, then your SNR is 100/SQRT(100 + 180 + 5 + 50 + 1.56^2), so your noise grows from 10e- to 18.37e-, or approximately 84%. Your SNR drops from 10:1 to 5.4:1. That is rather significant. Assuming the glows and dark current and everything else grow linearly, then they grow along with the signal when you expose for 10x longer. So when you have 1000e- signal, you have 1000/SQRT(1000 + 180 + 50 + 500 + 1.56^2), so your noise grows from 31.62e- to 58e-. Your SNR drops from 31.62:1 to 17.24:1. Again, a very significant difference. Here's the kicker. Because the other terms can affect SNR so much (and this is the case with every sensor, they all have dark current, and we all usually have skyfog...many other cameras also have amp glows, particularly DSLRs and mirrorless cameras), you can actually get a similar result by stacking 10 of the shorter subs. In this case, because the read noise is so low, stacking 10 of the shorter subs gives us an SNR of 17.22:1. Almost exactly the same.

Edited by Jon Rista, 17 June 2016 - 10:25 PM.

### #68 Jon Rista

Jon Rista

ISS

• Posts: 24,854
• Joined: 10 Jan 2014

Posted 17 June 2016 - 10:30 PM

"You think I am getting skyfog limited subs with a 3nm Ha filter in a mere 300 seconds? Seriously?"

When you image, whatever noise there is, settles at the floor. If the magnitude of the dimmest detail is not sufficiently above that noise, then it will either be a haze or not visible at all.

You are thinking in terms of the limits of the whole image. I am only talking about the dimmest details in that image. Tolga was at f/6, I at f/7 and you at f/8. We exposed for 1 hour and 15 minutes, while you exposed for 3 hours. Also, you said that longer exposures didn't help. You have something blocking the dimmest detail.

I am just saying that the there is noise at the bottom that is obscuring the dim detail and the dim stars. Hiding it even.

I am assuming it is coming from the sky, since it isn't coming from the camera.

I am not talking skyfog limit, and even if you exposed to the skyfog limit, your total noise from all sources would be some number, say, 200 e-, then any detail close to that will be hazed, and less than that will be invisible.

Here are some FWHM measures from some subs. One of M101 from SharpCap (best sub I could find from that night, most of the data has been backed up onto BluRay disk and deleted), the other from M57 from SGP. I chose the best I could find from my M57 batch from two nights ago:

So, in one case I've got 3.7" FWHM and in the other I've got 0.46" FWHM. You guys seriously don't think that would affect signal strength?

I agree that noise is going to obscure faint details. Absolutely no disagreement there. I just don't think that noise is the primary reason my signal is faint. I think the reason the signal is riding the noise floor is it's been smeared around, so there is less signal in more pixels, rather than more signal in fewer pixels.

However the rim of the outer halo should be about 8 pixels across, and instead they are about 25 pixels across (Assuming I could get back to ~1.3" FWHM rather than 4" FWHM). Concentrate all the signal from those 25 pixels into 8 pixels. Instead of ~240 ADU per pixel, I'd have ~750 ADU per pixel. Or instead of 14e- per pixel it would be about 43e- per pixel. I mean, why does the SNR on a guide star jump from 4:1 to 30:1 or 50:1 or more when you go from a few steps off focus to dead on focus? It's signal gets concentrated onto fewer pixels.

As for f-ratio. It doesn't really matter all that much, although in both cases you guys had both larger apertures (more total signal) and faster f-ratios (quicker saturation per pixel per unit time). Aperture matters. Every pixel-sized point in the sky that falls within your field has it's light concentrated onto a pixel on the sensor, and the quantity of that light is determined by the aperture. The total aperture gathers light for each and every point. Your two scopes were gathering far more light per pixel. By more than a factor of two in the case of the 11" SCT. And you were binned 2x2, which at least doubles SNR again. (I really gotta get a darn EdgeHD! )

### #69 FirstC8

FirstC8

Apollo

• Posts: 1,118
• Joined: 10 Mar 2016

Posted 17 June 2016 - 11:57 PM

Ok think I have gained a lot about good signals vs all sorts of noises and bad signals.

Maybe the best approach to all the madness is to get a C11, Hyperstar and 1600 OSC, do shorts, not worry about all the noises and skip all the post processing.ðŸ˜†

### #70 Jon Rista

Jon Rista

ISS

• Posts: 24,854
• Joined: 10 Jan 2014

Posted 18 June 2016 - 12:02 AM

Ok think I have gained a lot about good signals vs all sorts of noises and bad signals.

Maybe the best approach to all the madness is to get a C11, Hyperstar and 1600 OSC, do shorts, not worry about all the noises and skip all the post processing.

I'm pretty much following the same plan. I'm ditching the RC now. I'm seeing weird stuff happening with the stars with AF-length exposures of around 10s. I can't seem to get it focused well at all. And I've now lost five nights to fiddling with it.

I'm moving back to the 600mm lens. I'm finishing up the DIY focuser I built for it tonight, will test it tomorrow, hopefully it will have a small enough step size to get me well focused, and I'll move back to the wide field stuff. I prefer that anyway. I guess I just had to know for sure that the RC wasn't for me, and there isn't a shadow of a doubt anymore. Time to sell it!

I suspect an f/4 lens with a 150mm aperture will be pretty nice for NB, and 300s subs at unity might even be too long. I suspect you would have zero problems with either dark current or glows with a hyperstar...you should be truly read noise limited.

### #71 FirstC8

FirstC8

Apollo

• Posts: 1,118
• Joined: 10 Mar 2016

Posted 18 June 2016 - 12:17 AM

Now that you mentioned NB, once I tried NB, can't go back. So all the sudden Hyperstar is no longer ideal setup.

### #72 Jon Rista

Jon Rista

ISS

• Posts: 24,854
• Joined: 10 Jan 2014

Posted 18 June 2016 - 01:27 AM

Now that you mentioned NB, once I tried NB, can't go back. So all the sudden Hyperstar is no longer ideal setup.

Because of spectral shift? You can get around that with wider band filters. A 3nm filter would likely not work, a 5-6nm might, but with diminished transmission (AstroDon's would be best as they are designed to). Wider than that, it should work.

### #73 FirstC8

FirstC8

Apollo

• Posts: 1,118
• Joined: 10 Mar 2016

Posted 18 June 2016 - 02:20 AM

Now that you mentioned NB, once I tried NB, can't go back. So all the sudden Hyperstar is no longer ideal setup.

Because of spectral shift? You can get around that with wider band filters. A 3nm filter would likely not work, a 5-6nm might, but with diminished transmission (AstroDon's would be best as they are designed to). Wider than that, it should work.

That and the over all difficulty using NB filters with Hyperstar.

I was planning to get Hyperstar and use my Minicam5F with it for a while. It has built in filter wheel and 12nm filters, it has a small enough footprint Starizona actually has it on their list of cameras for hyperstar.

But the camera is giving me grief, too much noise and glow to deal with.

### #74 William Mc

William Mc

Mercury-Atlas

• topic starter
• Posts: 2,783
• Joined: 24 Nov 2007

Posted 18 June 2016 - 04:45 AM

I was going to dive into a few of my books tonight as a refresher on S/N, but I found the current discussion here as good or better!  Some of the best CN content I've seen. Thanks all and well done.

Question  -  How does the ASI1600  glow intensity compare to other popular sensors?  Is the issue unusually bad for this particular sensor? I know the earlier Canon DSLR's had some nasty glow issues.

Edited by Wmacky, 18 June 2016 - 06:36 AM.

### #75 TimN

TimN

Soyuz

• Posts: 3,651
• Joined: 20 Apr 2008

Posted 18 June 2016 - 06:41 AM

Unlike DSLRs this camera is cooled. I wonder if the amp glow is not a thermal glow? I've assumed that most DSLRs glows are thermal based.

## Recent Topics

 Cloudy Nights LLC Cloudy Nights Sponsor: Astronomics