Jupiter 10/19/22 - mak 150

[Sinemetu63]

Jupiter and Io with shadow transit.

CS

Paolo

[yelsac]

Excellent, what camera are you using?

[Sinemetu63]

Hi Yelsac, thank you.

I'm using a ZWO ASI 385 MC with a 1.5x barlow, although the focal lenght is probably more than the nominal 2700 mm it should be, probably around 3400 mm, so sampling rate should be around 5 pix or 6 pix per arcsec.

The picture had a ROI of 640x480 but was resized to 120%.

Edited October 21, 2022 by Sinemetu63

[vlaiv]

Great image.

You should probably keep barlow within the specs to give you x1.5 as this is over sampled.

Frequency analysis says that you are x1.5 over sampled. From Jupiter disk size - image as is resolves to ~4200mm (or ~3500 if you resized it to 120%), But actual focal length should be ~x1.5 lower than this - or 2800.

By the way - this is first time I've seen someone capture so very close to theoretical limit of resolution.

Theoretical limit of resolution is where F/ratio is x5 pixel size - in this case 3.75um pixel size of ASI385 will be F/18.75 or FL of 2812mm.

Image has data that corresponds to about 2800mm of FL - so you manage to capture all there is really.

Here is Fourier transform / spectrum of image:

You can clearly see that signal is concentrated in inner 66% :

Rest is just noise.

If we remove outer part of FFT like this:

And do inverse FFT - we get the same image:

(minus some noise)

By the way - properly sampled image looks like this:

Very sharp and detailed when viewed at 100% (at least on my computer screen)

 

[Avocette]

@vlaiv I am also using a Mak 150 with a 2” ED Barlow, although mine is nominally 2x. From platesolving an image in Cygnus with plenty of stars, the focal length measures 3445mm. This gives an 11 arcminutes square FOV for my 11mm square ASI533MC with 3.76μ pixels. Hence sampling at 11x60/3008 or 0.22 arcsecs/pixel. 
I have read and reread your post but am struggling somewhat to understand the detail of your analysis! Of course I would like to think that I am achieving similar results to @Sinemetu63 who coincidentally is just the other side of the Alps from me. However I have always been unsure of my telescope which I found was out of collimation when I bought it third or fourth hand, with a collimation screw loose! I attempted collimation over the years and most recently with the encouragement of posts from @Captain Scarletam starting to feel it is pretty well aligned. I have also recently started to realise just how influential the seeing is to the results I obtain. I would be happy to post some of my recent results if you could be interested to apply your analysis tools? 

[Avocette]

Here is one of my Jupiter captures from 6 October. This is the 1.5x drizzled output from AS!3 via Registax 6.

[vlaiv]

1 minute ago, Avocette said:

I would be happy to post some of my recent results if you could be interested to apply your analysis tools? 

Sure. I'd be happy to do that, and even explain the tools - it is just ImageJ which is open source scientific image analysis tool (made for microscopy, but works well for astronomy - there is even astronomy version called AstroImageJ that has photometric measurement and plate solving added - but I prefer to use plain version loaded with plugins for this sort of thing).

2 minutes ago, Avocette said:

I have read and reread your post but am struggling somewhat to understand the detail of your analysis!

I can explain things in detail if you wish.

It is basically fourier transform to examine image in frequency domain.

Atmosphere and telescope act on image in similar way as equalizer acts on sound.

There are different frequency components to the image - similarly as there is to sound. We know difference between low and high frequency sound by feel (or hearing rather), but in image it is coarse and fine detail of sorts (low frequency is coarse detail and high frequency is fine detail).

Blurring removes or rather attenuates that high frequency detail in the image.

If you've ever seen MTF of telescope - which looks like this:

That is exactly like equalizer above - as we move from left to right we move from lower to higher detail. This graph shows how much attenuated or muted are frequencies or detail. To the left there is almost no attenuation - but as we move to the right - finer detail is muted more and more - until we reach 0.

This is resolving power of telescope / aperture - point after which any smaller / finer detail is simply completely blocked.

Ideal sampling rate is when this curve hits zero at the edge of image in frequency domain. That is related to Nyquist sampling theorem and corresponds to two pixels per wavelength / cycle.

In 2d image - above graph is actually a cone centered in the center of the image and slowly falling towards the edges.

Here is surface plot of FFT of that properly sampled jupiter image:

You can see similar shape that is high in center and falls towards the edges. You want that falloff to hit the edge for properly sampled image.

Regarding the sampling - here is what can happen:

Top diagram shows over sampling. In each diagram we capture all the data between black and red vertical lines (black is just origin and red is highest frequency that we capture depending on our sampling frequency - pixel size with respect to "magnification").

In first case of over sampling we capture all the data but we also capture "empty" part where there is no signal. On its own - that is not a problem. Problem here lies in the fact that in order to over sample - we must use too much "zoom" and that spreads light over more pixels than necessary thus reducing signal per pixel and overall SNR (which is needed for sharpening stage and so on). This also means that we need longer exposures to get good SNR and we always try to get the lowest possible exposure length when doing planetary to freeze the seeing.

Second case is proper sampling - we capture all there is - no more, no less. Perfect case.

Third case is under sampling. We simply clip some of the data that is there. Under sampling does not have many drawbacks except that we loose some of the detail as we did not "zoom in" enough to capture it. There is also matter of aliasing, but that is separate topic and not something that one should worry about as high frequencies that alias are already very weak.

Just to add - black line is center of FFT image and red line is edge of FFT image in 2d case.

Hope this somewhat explains what I've done and things behind all of this.

 

[vlaiv]

14 minutes ago, Avocette said:

This is the 1.5x drizzled output from AS!3 via Registax 6.

Ok, so my first impression is that you were using way too long exposures for that capture.

@Sinemetu63 sorry we are having this off topic discussion on your lovely Jupiter image post (I at least hope it will be beneficial to people).

You are probably looking at histogram and using brightness of the view to determine exposure length. That is wrong approach. Exposure length should be set to correspond to coherence time. We can't know that as it is variable that depends on local conditions at the time of recording - but it is usually 5-6ms for aperture sizes amateurs use. You should set exposure length to 5ms even if image looks too dark (you can go to 10ms if seeing is exceptional and there is almost no movement in atmosphere).

If I do analysis - two things are apparent:

First is "artificial" enlargement of image by drizzle. Any time we use rescaling of the image there is different signature - which is square. Aperture leaves round signature in frequency data.

Other thing we see is that signal is constrained in very center - which means that data is over sampled by a large margin.

For ASI 533 pixel size is almost the same as ASI385 - so optimum F/ratio is the same F/18.6 - which means 150 * 18.6 = ~2800mm

If you used x2 barlow and then drizzled x1.5 - you effectively raised F/12 to F/12 * 2 * 1.5 = F/36

That is x2 more than you need. From above FFT - things are even worse as there is hardly any signal past initial 20%, so image is over sampled by factor of x5.

Part of this is due to over sampling with respect to aperture size and the rest is simply due to seeing (seeing lowers max theoretical resolving capability of scope further) - and in particularly longer exposures.

 

[Sinemetu63]

3 hours ago, vlaiv said:

Great image.

You should probably keep barlow within the specs to give you x1.5 as this is over sampled.

Frequency analysis says that you are x1.5 over sampled. From Jupiter disk size - image as is resolves to ~4200mm (or ~3500 if you resized it to 120%), But actual focal length should be ~x1.5 lower than this - or 2800.

By the way - this is first time I've seen someone capture so very close to theoretical limit of resolution.

Theoretical limit of resolution is where F/ratio is x5 pixel size - in this case 3.75um pixel size of ASI385 will be F/18.75 or FL of 2812mm.

Image has data that corresponds to about 2800mm of FL - so you manage to capture all there is really.

 

Hi Vlaiv, thank you and very interesting analisys.

Actually the 3400mm estimate wasn't mine but suggested by another imager who made this calculation on the basis of a previous image of mine shot with the same parameters. 

I bought the 1.5x barlow just to get close to the "ideal" sampling rate of 5x as suggested by many.

Below you can see 2 pictures taken in the days before that were not resized to 120% .

Can you verify the correct focal lenght and judge the image quality?

CS

paolo

Edited October 21, 2022 by Sinemetu63

[vlaiv]

10 minutes ago, Sinemetu63 said:

I bought the 1.5x barlow just to get close to the "ideal" sampling rate of 5x as suggested by many.

Yes, x1.5 barlow is just the right.

F/ratio should be x5 (at max) of pixel size - so in your case that is 3.75 x 5 = F/18.75

Since scope is F/12 if you add x1.5 barlow you'll get F/18 - which is just right.

Thing is - you should verify if scope is actually operating at F/18, because both barlow and mak can change that. Magnification of barlow element depends on distance of element and sensor - more distance - more magnification it will give you.

Maksutov scopes focus by shifting primary mirror - and actual focal length depends on distance between primary and secondary mirror. This means that as you focus - you also change focal length of instrument (and hence F/ratio). This is by small factor but it still happens. With camera this is noticeable as we usually remove 1.25" diagonal when we attach camera and that is often as much as 70-80mm of back focus difference.

In any case, here is calculation for effective F/ratio in this image

Diameter of Jupiter in the image is ~222px in the image.

Currently, apparent diameter of Jupiter is ~48.7 arc seconds - and that makes it ~0.21937"/px

Given that your camera has 3.75um pixel size - focal length is 0.21937 = 3.75 * 206.3 / FL => FL = 3.75 * 206.3 / 0.21937 = ~3527mm

F/ratio is then 3527 / 150 = F/23.5

Even in this image that is not resized - it seems that magnification of barlow is too high. It is operating closer to x2 than to x1.5.

You are over sampled by factor of 23.5 / 18.75 = ~1.25

In another words - if we do FFT - "circle" of signal should end up somewhere around 80% towards the edge. Let's check that.

Here it is a bit harder to see - as there is seeing influence present that restrict most of the data to central part - but there is "secondary" ring of data. Let me do profile plot to see if we can figure it out:

That still does not help much - I marked where telescope optics is - I think that due to seeing data probably ends a bit earlier - so in this instance seeing prevented you to get max resolution of telescope and actual detail effectively stops between 300 and 400 mark in the graph.

In any case - if you want to hit optimum sampling - you should adjust barlow / sensor distance to get x1.5 magnification.

You can do this during daytime - aim scope at distant target that you can easily measure - like apartment building or bridge - something with clean lines.

Take image without barlow - and measure some distance.

Then add barlow but put variable spacer between barlow and sensor - and take images while changing the distance - each time measure same feature until you get it to be x1.5 longer than in baseline image.

[vlaiv]

Here is FFT of that other image added later in post:

It again shows that faint ring - maybe a bit better than the last one.

If I run mean filter on it to remove some noise - maybe it will be easier to asses where edge of the data is.

Ok, I've done a trick - I increased brightness / contrast until circle is readily detectable:

Placing cursor on the edge says it is 2.64 pixels per cycle (ideal sampling rate is 2 pixels per cycle).

Difference therefore is 2/2.64 = ~0.76

We calculated above that theoretical max is 0.8 at this sampling rate so data falls just a bit shorter than this theoretical max.

[Avocette]

Thanks @vlaiv and  @Sinemetu63 for the ongoing discussions! I have reprocessed completely my earlier image to avoid the 1.5x drizzle, but I feel I lost a lot of important detail that was evident in the earlier version. I had understood that x1.5 drizzle processing in AS!3 was not just rescaling.

 

[vlaiv]

Just now, Avocette said:

I had understood that x1.5 drizzle processing in AS!3 was not just rescaling.

There are two types of drizzling.

First is bayer drizzle that AS!3 uses to debayer data and it does its job well. It makes OSC camera have the same resolution as mono camera. This is related to pixel size and fact that color pixels are in fact spaced more than in mono version (every other pixel is red and every other is blue - green similarly but it is "more dense").

This is done by default for OSC data and you don't have to turn on anything.

Second is "regular" drizzle.

This kind of processing requires under sampled data and in my view - it is questionable if it works at all.

It was designed for Hubble where very precise sub pixel dithering can be employed - and there was no atmosphere to mess up things.

In either case - drizzle simply won't do anything useful to your data as you are not (even if it works) - as you are over sampled to start with and not under sampled.

4 minutes ago, Avocette said:

but I feel I lost a lot of important detail that was evident in the earlier version

In this case that solely depends on processing. Data is not processed in the same way.

Wavelets are done more conservatively in this second go.

Maybe if you post 16bit raw stack - without any sharpening done (not even in AS!3) for people to process. Maybe there is more to the data.

Many experienced imagers sometimes extract more from the same data in terms of detail than I able to, so there is certainly that factor.

[Avocette]

Yes - I was probably too quick and used the same Rx6 parameters.

 

Edited October 21, 2022 by Avocette

[CraigT82]

44 minutes ago, vlaiv said:

Diameter of Jupiter in the image is ~222px in the image.

Why do you measure the disc diagonally like that? I’ve always done it across the equator as I assumed it was equatorial diameter that was given for Jupiter’s apparent size in arc seconds but I may be wrong. Jupiter is quite strongly oblate so it may make a difference? 

[CraigT82]

5 hours ago, Sinemetu63 said:

Jupiter and Io with shadow transit.

CS

Paolo

A fine image from a 6” mak. Cracking scopes they are. 
 

[vlaiv]

1 minute ago, CraigT82 said:

Why do you measure the disc diagonally like that? I’ve always done it across the equator as I assumed it was equatorial diameter that was given for Jupiter’s apparent size in arc seconds but I may be wrong. Jupiter is quite strongly oblate so it may make a difference? 

No particular reason - I just take a point (near equator) - where edge is nicely defined and then drag line to other side trying to get it to go thru center.

I don't think that odd pixel here and there will make much of a difference.

According to what I've found difference seems to be 142984 - 133708 / 142984 = ~6.5%, so that is quite significant in general terms, but on 222px image that will be 14px depending on direction of measurement?

(that is a bit more than I thought it would be)

Yep, further measurement confirms that there is about 14px of difference

However - I was not too off with 222 measurement (maybe 2% error).

 

[Pete Presland]

Very nicely captured.

[Sinemetu63]

A short animation of the 10-19-22 captures.

All original size 640x480.

CS Paolo