Telescope Guiding Techniques: Improving Exposure Times through Auto Guiding

[russ_watters]

This is a screen shot from the PHD auto-guider software, you can see how a single star is selected

View attachment 213449

the lower wriggly lines are showing tracking error and corrections

there's a zillion sites/instructional videos etc available

https://www.bing.com/search?q=PHD+g...smsn=1&refig=57e96c1611f34526bd2176f00c456152

Dave

Dave, is this your actual setup? Can you describe your equipment? I have some questions...

 

[Nidum]

My university once had a small observatory located on some uplands a few miles inland from the main campus .

The telescope there just had a clockwork drive mechanism for keeping the telescope locked on to whatever was being observed . Apparently worked perfectly well for shorter observation periods .

 

[Nidum]

My digital camera has image stabilisation . This compensates for quite high levels of shake when pressing the shutter button .

I don't know how it actually works but all the digital processing required is done in real time by the one processor chip in the camera .

Compensation for shake and compensation for uniform motion can't be much different ?

 

[sophiecentaur]

If you can't find one star bright enough to guide off in a field of view, then you certainly can't find more than one to guide off in that field; because you can't see them. That's the entire point of the problem!

That sounds to me the sort of argument that would tell you that Stacking is not worth doing. A single star that is hardly visible becomes visible when you stack many images. There is precisely the same advantage (in a different context) from looking at all the available point sources and adding their effects together to locate the position of the whole image. The relative positions of the stars are not likely to change over the period of the feedback intervals so you compare one image with the identical image - shifted by a very small number of pixels (the seeing would be a problem as always). SNR improvements can be very significant in threshold conditions. This is a process that is bordering on the trivial in the context of compression of moving pictures. It is the sort of thing that allows very good slo-mo to be taken from normal frame rate TV pictures.

I am not qualified to comment on how 'worth while' improved tracking over what one guide star will give you in the context of today's equipment and practices but I can comment on the fact that the whole image (which you guys refer to as 'many guide stars' for only historical reasons) contains more positional information than the image of a single guide star. The human brain does processing along precisely the same lines as the system I am suggesting. When you are involved in sport , hunting , fighting etc. You assess the whole of your visual image to guide your motions. If you were playing football in the dark and players and ball were lit with a single bright LED (which is the equivalent of single guide stars) then performance would be much worse. That is pretty obvious.

My university once had a small observatory located on some uplands a few miles inland from the main campus .

The telescope there just had a clockwork drive mechanism for keeping the telescope locked on to whatever was being observed . Apparently worked perfectly well for shorter observation periods .

The operative phrase here would be "perfectly well". There are many smashing pictures from that era. Modern AP is much more fussy about results than AP used to be. Open loop operation can't deal with the imperfections of even a high quality mount. For mobile operations the situation must be even more in need of extra electronic help.

I suggest that like most things, it is best to learn how things are done now and why before trying to reinvent the wheel.

You make an important point there and I am aware of it. However, the justification for not considering something should never be 'because we have always done it this way'. I am not suggesting just "guiding off multiple stars".
I notice that Very Large telescopes use a Laser guide star to achieve the best guiding. That is new and has brought terrestrial viewing into the new class of quality.

 

[davenn]

I notice that Very Large telescopes use a Laser guide star to achieve the best guiding. That is new and has brought terrestrial viewing into the new class of quality.

I think you mis-understand the purpose --- has nothing to do with "providing the best guiding"
1) to provide a point of light as a guide point where there is a lack of stars near the object that needs to be imaged
2) to be used as a measure of the atmospheric instability so that the adaptive optics can adjust the star/object of interests'
light to be the most stable for imaging of it

This isn't overly new, adaptive optics have been around for many years. The 2 x 10 metre Keck scopes on top of Mauna Kea in Hawai'i, that I visited in 1999, have been using adaptive optics since their first lights in 1993 and 1996Dave

 

[sophiecentaur]

I think you mis-understand the purpose --- has nothing to do with "providing the best guiding"

No I didn't mean that. I meant that doing things a different way can be an improvement when technology allows. Adaptive Optics has advanced a bit since the 90's.
I accept your doubts about the present viability of my idea but I would really like to think you appreciate what's actually involved and the advantages the could be gained which are significant.

 

[Drakkith]

That sounds to me the sort of argument that would tell you that Stacking is not worth doing. A single star that is hardly visible becomes visible when you stack many images. There is precisely the same advantage (in a different context) from looking at all the available point sources and adding their effects together to locate the position of the whole image.

No it's not. The problem is that, if you can't find one single star to guide off of in your FOV, you don't have any point sources at all! None that you can guide off of at least. They're all buried in noise! Your software (and your eyeball) would not be able to reliably tell where the centroid of each star is between successive exposures because the noise causes the pixel values to fluctuate too much. In most cases you wouldn't even be able to see that there's a star there at all.

Stacking exposures to produce a high SNR image works because the process is essentially identical to taking a longer exposure, just with noise added in from the thermal and readout noise of each exposure. But stacking only works if you know that your exposures are all pointing in the same direction to within a pixel or two. Trying to look at many different sources within the same exposure doesn't work if your sources are all buried in noise. And you can't sum different sources together to somehow increase the SNR of the image either (well, you can, but only by binning the pixels together and losing most of your resolution).

The relative positions of the stars are not likely to change over the period of the feedback intervals so you compare one image with the identical image - shifted by a very small number of pixels (the seeing would be a problem as always).

The problem is that you don't know the position of the stars in the first place. Pixel 5436 might be higher in one exposure, and then, because of noise, pixel 5442 might be higher the next exposure, regardless of where the centroid of the star is actually located on the sensor. It might be at pixel 5436, it might be at 5442, or it might be at another nearby pixel. You don't know because the noise is causing such a large fluctuation (large relative to the star's signal) at each pixel between exposures.

SNR improvements can be very significant in threshold conditions. This is a process that is bordering on the trivial in the context of compression of moving pictures. It is the sort of thing that allows very good slo-mo to be taken from normal frame rate TV pictures.

What SNR improvements? What are they actually doing in those slow-mo videos to increase the SNR? What techniques are they using?

I am not qualified to comment on how 'worth while' improved tracking over what one guide star will give you in the context of today's equipment and practices but I can comment on the fact that the whole image (which you guys refer to as 'many guide stars' for only historical reasons) contains more positional information than the image of a single guide star.

Okay. Provide some evidence for this please. What exact method would be used? How does it work? How does it apply to auto-guiding? And I'm not asking for a hand-wavy explanation, I'm asking for something more concrete, preferably with some formulas to back it up if you have them.

The human brain does processing along precisely the same lines as the system I am suggesting. When you are involved in sport , hunting , fighting etc. You assess the whole of your visual image to guide your motions. If you were playing football in the dark and players and ball were lit with a single bright LED (which is the equivalent of single guide stars) then performance would be much worse. That is pretty obvious.

I'm sorry but I don't see any connection between your analogy and autoguiding. You start out by saying that you're evaluating the entire FOV for positional information, which to me implies that the players and ball are "stars", but then you switch and talk about illuminating them all with a single LED, with the LED being a single guide star. Or, as just occurred to me, did you mean to say that each player and the ball had a single LED on them, so that you'd see two dozen points of light when looking out over the field?

You make an important point there and I am aware of it. However, the justification for not considering something should never be 'because we have always done it this way'. I am not suggesting just "guiding off multiple stars".

I cannot fathom how you came to the conclusion that our arguments were based on the notion that we've always done autoguiding this way and it shouldn't be changed. No. Our arguments are based on our own understanding of how autoguiding and imaging work. Now, it's always possible that we might somehow be wrong in the end, but that does not mean we don't have valid reasons for arguing against your idea.

As for your idea not being about guiding off of multiple guide stars, I'm afraid I can't see any other way to describe it. You've looking at multiple stars in the image and getting positional information from them to calculate correction signals for the mount, right? And this is done by looking at the position of multiple stars in the FOV between successive exposures, right? But that's exactly how autoguiding off of a single star works. You might be thinking of throwing in some other processing techniques into play, but the core idea is still about guiding off of multiple guide stars, is it not?

I accept your doubts about the present viability of my idea but I would really like to think you appreciate what's actually involved and the advantages the could be gained which are significant.

We can't appreciate what's involved if we don't think it's a valid idea. That doesn't mean we're not listening to you or that we aren't open to your idea being valid, we just don't think it is.

 

[davenn]

Adaptive Optics has advanced a bit since the 90's.

yes they have, but the principle is still the same, maybe just refined a bit
The amount of "twinkle" instability of a stars light (caused by atmospheric distortion) is measured and signals are sent to the multiple of
mirrors of the telescope to continuously adjust them to remove as much of that instability as possible

Adaptive optics has nothing to do with guiding ! It is all about producing a stable image for the camera

D

 

[Drakkith]

Adaptive optics has nothing to do with guiding ! It is all about producing a stable image for the camera

I like to think of adaptive optics as just guiding on steroids. Both have the objective of keeping the centroid of the star at the same location on the sensor. Guiding just doesn't get rid of all that atmospheric blur.

 

[davenn]

Guiding just doesn't get rid of all that atmospheric blur.

exactly, which is why there are adaptive optics

 

[sophiecentaur]

. Or, as just occurred to me, did you mean to say that each player and the ball had a single LED on them, so that you'd see two dozen points of light when looking out over the field?

Yes. That's what I meant. Identifying a player and following him is not as easy if there's just one led on him. The total player image is what your eye makes best use of. This is particularly relevant, of course, when the lighting is low (low SNR). This is precisely what the processing in my idea would do.
I am disappointed that the reaction to this idea has been so negative. Stacking is just a way of using information gathered over an extended time. My idea is merely gathering more relevant information by looking over an extended spatial region; there is a direct correspondence. The process is not as obvious as the way a single star guider uses. The conventional guider can only work for a star that is bright enough for a simple algorithm to work. Is it not obvious that there is far more information available in the total field than in the position of just one star? You do not need to identify the positions of individual stars and work on each individually. It is the whole field that would be processed.

Okay. Provide some evidence for this please. What exact method would be used? How does it work?

Here's one method that would work. Take the 2D correlation function between successive captured frames. That will give a peak which corresponds to the displacement vector. For a totally random distribution there will be a single peak and there can be other subsidiary peaks where there are regular features in the sky but the sky has no repeating patterns so the correlation peak would be high. You can take frames at as high a rate as your correlation process time and camera noise will allow. Noise and hot cells would be addressed in the same way as normal imaging is treated.

 

[Drakkith]

Yes. That's what I meant. Identifying a player and following him is not as easy if there's just one led on him. The total player image is what your eye makes best use of. This is particularly relevant, of course, when the lighting is low (low SNR). This is precisely what the processing in my idea would do.

I actually disagree somewhat. A single LED per player would make identifying his position in low-light levels easy compared to having no LED on the player. But I can't see how this applies to autoguiding. All we're worried about is making sure our telescopes stay pointed at the same location in the sky where the objects are all moving at approximately the same rate. Having point-sources makes guiding extremely easy compared to trying to identify extended, diffuse sources. And since the stars are indeed the closest thing to a point source we can get, having even one of them gives you all of the pointing accuracy you will need.

I am disappointed that the reaction to this idea has been so negative. Stacking is just a way of using information gathered over an extended time. My idea is merely gathering more relevant information by looking over an extended spatial region; there is a direct correspondence.

There may be a correspondence, but I don't agree that it is useful for autoguiding. Autoguiding takes place in an environment of very low signal, where you often cannot rely on finding more than one or two good sources to guide off of without increasing the exposure time for your guider to 20+ seconds, which is starting to be too long to effectively correct for mount errors. So unless you're pushing the limits of this timeframe, you're unlikely to be able to use more than one or two sources.

The process is not as obvious as the way a single star guider uses. The conventional guider can only work for a star that is bright enough for a simple algorithm to work. Is it not obvious that there is far more information available in the total field than in the position of just one star? You do not need to identify the positions of individual stars and work on each individually. It is the whole field that would be processed.

I'm not following you. If you can't even see the stars because they're buried in the noise, how can you possibly use them for anything?

Here's one method that would work. Take the 2D correlation function between successive captured frames. That will give a peak which corresponds to the displacement vector. For a totally random distribution there will be a single peak and there can be other subsidiary peaks where there are regular features in the sky but the sky has no repeating patterns so the correlation peak would be high. You can take frames at as high a rate as your correlation process time and camera noise will allow. Noise and hot cells would be addressed in the same way as normal imaging is treated.

Normal imaging addresses noise by increasing the exposure time or combining exposures to increase the SNR of the image. Neither of these are good options for an autoguider, as they both increase the time between mount corrections.

My basic objection is that I don't see how your method is effective in the extremely low light levels associated with the night sky. As I've asked already, how can you use stars that you can't even see? You have no correlation if the noise is overwhelmingly dominant, and the only way to get the stars' signals out of the noise is to either increase the exposure time or combine images. Both of which are only useful if you can do so while keeping your mount corrections coming at intervals of at most 10-15 seconds, 15-20 if your mount is above average.

Beyond that I still don't see how this is any more useful to astrophotographers than guiding with a single star. Accuracy certainly isn't increased since the autoguider itself is not the limiting factor, as we've explained earlier. And since that appears to be the main appeal of your method, I can't understand why you're still pushing it as a viable and useful method.

Please note that I'm not saying your method flat out doesn't work. Of course it works. It's just that, in my opinion, there's nothing to be gained by using it. It's neither more accurate, less costly, or easier to use than guiding with a single star as far as I can tell.

 

[davenn]

Stacking is just a way of using information gathered over an extended time.

which is great for imaging an object ... it works well
But as Drakkith said, it's useless for guiding because of the time lag
and as long as there is a star, you can guide on it

My idea is merely gathering more relevant information by looking over an extended spatial region; there is a direct correspondence

what more relevant info ??

again, the screenshot from the guiding software ...

You can see multiple stars in that shot, you can select whichever one you like.
It doesn't matter which one you select as they are all moving with the same motion(s).
There are 3 motions...
1) the natural motion across the sky
2) motion that strays from that natural motion because of the mount polar alignment inaccuracies ( I have already mentioned this in a previous post)
3) motion that strays from that natural motion because of the mount motor gear meshing inaccuracies ( I have already mentioned this in a previous post)

Guiding, be it auto or manual, aims to remove the inaccuracies of #2 and #3 so as to get the scope tracking the natural motion of the guide star (and as a result, the object being imaged) as accurately as possible

The process is not as obvious as the way a single star guider uses. The conventional guider can only work for a star that is bright enough for a simple algorithm to work. Is it not obvious that there is far more information available in the total field than in the position of just one star?

yet again, you don't need multiple stars when just one star provides the same amount of info, there is nothing to be gained by tracking multiple stars
If there was, I'm sure they would be doing it already. Auto-guiding has been around long enough for that to be realized if it was of use.D

 

[sophiecentaur]

But as Drakkith said, it's useless for guiding because of the time lag

Of course it is and I am not suggesting it as a method.

Stacking is just a way of using information gathered over an extended time. My idea is merely gathering more relevant information by looking over an extended spatial region;

I thought this made it clear. You have to think 'inside out' to get what I mean, perhaps but you must see that integrating over space and integrating over time will both improve results. Stacking is over time and this method is over space. It is particularly suitable for use with stars as they have identical profiles so the correlation peak would be very well defined.

yet again, you don't need multiple stars when just one star provides the same amount of info

and yet again I say that it isn't "multiple stars"; it's just a whole image.

Please note that I'm not saying your method flat out doesn't work.

That's reassuring.

No. Our arguments are based on our own understanding of how autoguiding and imaging work

I get that. But you are hanging your argument on the present system.

the only way to get the stars' signals out of the noise is to either increase the exposure time or combine images.

Which is precisely what the correlation method would do - images from all over a single exposure.

I'm not following you. If you can't even see the stars because they're buried in the noise, how can you possibly use them for anything?

Stacking achieves just that and so does the correlation method (in the spatial domain and not the time domain)

Adaptive optics has nothing to do with guiding ! It is all about producing a stable image for the camera

Both processes aim to deliver as much of the incident light from a particular part of space as possible on one particular part of the sensor. How can you separate them in any discussion? The systematic deviations of pointing, due to the mechanics of the mount can be reduced more and more by increasing the SNR of the guide control loop because an 'intelligent' loop can learn and precorrect. Backlash is harder to deal with, of course. You have repeated that the present guiding system is good enough but is it? The earlier post about clockwork tracking being 'quite good enough' may be worth re reading.
There a definitely two halves to this thread. The theory is something I feel confident about but, of course, present practice is not in my experience (yet). I totally accept how the present system is preferable because it is: 1. Easy to understand and explain: 2. available to buy. These make the choice pretty well a no brainer at the moment. But PF discusses many blue skies ideas and I hold that there are better guiding methods possible and probably on the horizon. (No puns intended here.)

 

[Gaston Baudat]

My idea is not just multiple guide stars as much as using the whole image ( whatever it happens to be) in the field of the scope. Not the same process as tracking a single star.
If you bear in mind that the electronics inside of most electromechanical systems gets ever cheaper, the processor would be a small fraction of the cost, once it’s an established system.
There is always a time during which new things are ‘objects to aspire to’ (an Apple phrase). After a while they are cheap and everyone has one.
A bit of a pipe dream, I agree but I am putting forward an idea not a piece of kit (yet). It would increase accuracy and make life easier.

My name is Dr. Gaston Baudat from Innovations Foresight, the company which makes the ONAG.
We are in the process of releaseing a new software for auto-guiding and real time auto-focus (focusing while the imager shutter is opened) using the all guider frame.
This patent pending approach leverages the ONAG large guider FOV at once (APS-C size chips, up to 28mm in diagonal) and the ever lower cost CMOS cameras.
With proper statistical processing one can extract image registration information for auto-guiding purpose even under SNR values down to 0dB.
In this approach one does not need, nor look for a specific guide star(s), but instead for any astronomical structures (if fact it works with any image, not only astronomical ones).
No assumption are made about the image, only about the noise floor for extracting the relevant information from the background.
You can see a demonstration in The Astro-Imaging Channel (TAIC) presentation I gave few months back on this topic:

 

[sophiecentaur]

My name is Dr. Gaston Baudat from Innovations Foresight, the company which makes the ONAG.
We are in the process of releaseing a new software for auto-guiding and real time auto-focus (focusing while the imager shutter is opened) using the all guider frame.
This patent pending approach leverages the ONAG large guider FOV at once (APS-C size chips, up to 28mm in diagonal) and the ever lower cost CMOS cameras.
With proper statistical processing one can extract image registration information for auto-guiding purpose even under SNR values down to 0dB.
In this approach one does not need, nor look for a specific guide star(s), but instead for any astronomical structures (if fact it works with any image, not only astronomical ones).
No assumption are made about the image, only about the noise floor for extracting the relevant information from the background.
You can see a demonstration in The Astro-Imaging Channel (TAIC) presentation I gave few months back on this topic:

Well how about that? Thank you for that contribution. It does the sort of thing I was suggesting and you have filled in a valuable gap for me because you have actual hardware and figures. I can imagine everyone having one in the too distant future.

 

[Gaston Baudat]

Well how about that? Thank you for that contribution. It does the sort of thing I was suggesting and you have filled in a valuable gap for me because you have actual hardware and figures. I can imagine everyone having one in the too distant future.

The idea behind is if one can guide on axis (and with a wide FOV) you are likely to guide on/near the actual target of interest.
Therefore by processing the whole guider (large diagonal guider chip) image we could extract registration (and rotational) information for auto-guiding, plus real time auto-focus information when using the astigmatism induced image on the guider side (no distortion on the imager side).

Of course it is a matter of SNR, as usually, if too low, for a given guider exposure time, there maybe not enough signal to guide, but so does a guide star based approach.
There is an old joke in signal processing community, "without any noise you do not need any signal".

The good news is that you usually image a target therefore there should be some signal, hopefully.
A key aspect here is to modelize the noise from a statistical stand point, not the actual mage content (like assuming any guide star(s)).
In a nutshell the algorithm at work is based on optimal filtering theory and can do a pretty good job even under low SNR values, depending of the target and the noise level, but of course there is limit, like any other approaches.

 

[Drakkith]

Stacking achieves just that and so does the correlation method (in the spatial domain and not the time domain)

With proper statistical processing one can extract image registration information for auto-guiding purpose even under SNR values down to 0dB.

Then I am apparently entirely unfamiliar with spatial correlation techniques. Do you have any good links or references?

Of course it is a matter of SNR, as usually, if too low, for a given guider exposure time, there maybe not enough signal to guide, but so does a guide star based approach.
There is an old joke in signal processing community, "without any noise you do not need any signal".

Do you know the SNR limit for this technique compared to that of a normal guide star based approach?

 

[Gaston Baudat]

The noise modelization is the heart of the technique and related to Kalman filter theory, you can find more on this here:

https://en.wikipedia.org/wiki/Kalman_filter

The image registration using spatial correlation is relatively straightforward when the first step mentioned above, extracting signal from the noise floor, has been achieved, Here is some basic math about digital image correlation:

https://en.wikipedia.org/wiki/Digital_image_correlation

Some aspects, let say some tricks and details of the implementation, are company confidential information, therefore I cannot elaborate more on those (sorry).

As far as comparison with other guide star centrodi based algorithms goes I could give rough numbers only, since it is a function of many parameters, such as how many guide star centroid are used in the same time (usually only one) .

I have not done very extensive/comprehensive tests but I would say that with one guide star and a centroid strategy you many need around 6 to 10 db of SNR, while the full frame technique work with SNR around 0 dB or so.
But this is not a totally fair comparison since the full frame technique uses all the image information at once (all the pixels), not any specific star (there is no need to locate/select any star manually or automatically).
Of course there are provisions for removing hot pixels, artifacts, ...

Basically each pixel is weighed against its level of information versus the noise floor (I should better say versus the noise model). A pixel with larger information content likelihood level contributes more in the registration calculation.
In short all guiding camera pixels are part of the process but with various weights related to their level of credibility for the task.

 

[Drakkith]

Thanks Gaston.

My apologies to @sophiecentaur. I was ignorant of my ignorance when it comes to spatial correlation.

 

[sophiecentaur]

@Drakkith : Unknown unknowns?
To be fair, I didn't explain it as well as Gaston did and he actually showed that it works. You would have needed to believe an old BS'er to take what I said as being gospel.
Your problem now is that you WANT ONE!

There is an old joke in signal processing community, "without any noise you do not need any signal".

Mr Shannon got it right all those years ago and he had virtually no hardware to test the idea.

 

[russ_watters]

I'm locking this thread for being overly speculative and will likely (when I get a more time later) split it between two threads:
1. A thread on real Telescope Autoguiding Techniques to help people improve their autoguiding (what the title of this thread implies it should be about).
2. A thread on this new full-frame guiding idea proposed by @Gaston Baudat

Please note:
The provided materials on the full frame guiding method are dense, yet and a bit thin in details (explicitly saying they won't go into the details). However, it appears to say that the secret sauce is in the noise filtration technique, not in the multi-star guiding technique. The objections raised to the idea of multi-star guiding previously and the circularity of the argument still appear to apply (if you can't track one star then you can't track more than one star, but what if you can track more than one star...?, you can't track one star, so you can't track more than one star). Anyway, frankly, the quality of the noise filtering appears to me to be too good to be true.

Multi-star guiding exists and has two advantages that don't apply to most people (which is why it isn't used much) or apply to the previous discussion:
1. If seeing is really bad and you have several guide star candidates of similar SNR you could jump back and forth between them as the SNR goes from "not enough" to "enough" for individual stars in successive frames. This would be a rare situation and when seeing is that bad you probably don't want to be imaging anyway.
2. It can allow you to correct for field rotation during imaging -- in only some cases and only if you have special equipment to rotate the camera while imaging. Field rotation is generally corrected after imaging, using a similar technique in software. It requires combining shorter exposure images.