21 replies to this topic

[andriy_melnykov]

Hi all,
I am not an experienced astrophotographer - I did some photos with DSLR cameras and camera lenses, and I also have an EAA setup, consisting of a 5 inch SCT + Alt-Az mount (Celestron AstroFI 125mm) and ASI385MC camera.
I was always fascinated, how much is possible with modern gear. And my interest was to push the possibilities to the limit, to take some challenge. As I learned about pulsars, my only thought was - I want to capture a pulsar. In the best case a slow-mo video! As I know, it was done before, but with gear like photon counting cameras and 0.5-1 m mirrors, or at least with decent gear, like 10-12 inch telescope on a good mount and stroboscope wheels.
So I have tried this, and had relative success. This cost me months of trying and frustration, and the result has a lot more post-processing, as I would define as a “fair” video, but all frames of the video are edited with the same procedure, and pulsation comes only from input data.
Gear used: AstroFi 125mm SCT with Alt-Az mount, ZWO ASI385MC, diy stroboscope with GPS synchronization, SharpCap, Python with astro libraries, DeepSkyStacker, Photoshop.
Crab Pulsar is very faint, around 16.5 apparent magnitude (and pulsing at 30 Hz, with two different pulses per period). To be able to resolve the pulsar from other stars I need to capture without any reducer. In this mode, the maximum I can get from my mount is 8 seconds of single exposure. I start to see the pulsar after about 3-5 minutes of stacking. And it is worse with a stroboscope, because there is less light. So, it was clear that the whole session will take many hours - this means, the stroboscope frequency must be very precise and stable during the session.
Stroboscope is a simple 3D printed construction consisting of a stepper motor, wheel with slots, and a case. Control has an Arduino board, GPS module and a stepper motor driver. SW in the Arduino implements a phase locked loop approach to synchronize a rotation of the wheel with the GPS 1 Hz signal, which is extremely precise and stable. The frequency of the stroboscope wheel is not exactly the frequency of the pulsar, but is slightly different, and the “window” is “sliding” through the pulsar period. The stroboscopic period was around ~80 seconds - so each 80 seconds we will see the same phase of the pulsar.
I have done some calculations, and the precision/stability of a simple quartz crystal generator really would not be sufficient in my case, that’s why the synchronization with GPS.
I did many sessions to try to let the system work, like it should. Finally, I have captured the data.
The whole capture took about 3 hours. I must delete many subs because of my mount tracking errors. Ended up with 755 subs, each 8 seconds exposure. Each sub has a timestamp (thanks to SharpCap SW). I suppose that internal PC time is precise enough for this task (under 1 second during 3 hours should be ok).
First Python script used the timestamps and sorted the subs, belonging to the same parts of the pulsar period - future 30 frames of a video. Here I must use a very precise frequency of the pulsar from a catalog, with a precision of one day - the pulsar frequency is actually slowly decreasing. Again, the calculations show that the frequency from one day off will not work! Actually, without a catalog the capture and post processing would not be possible. Second Python script did the interpolation of the pulsar frequency with day precision.
Each part of the period was stacked in DeepSkyStacker and aligned again, to have stars in the same positions on each frame.
Third Python script adjusted the brightness of each sub with photometry information from nearby stars. Each frame should of course have the same brightness of the stars, excluding the pulsar. I have also let the script calculate the brightness of the pulsar, and saw the pulsations for the first time.
After that was endless tweaking of the frames in photoshop, to see the pulsation clearly. I have prepared the first animation with 2 frames, without much filtering, and the second one where many stars are edited separately (like nebulas in astrophotography), but with the same procedure for each frame. This can be called cheating, but again, the pulsation results only from the input data.
I was actually very happy, when the challenge was over with some results.
But now I’m thinking about it again, to achieve better capture :-))))

 

I have saved the results and gear pictures at google drive, hope it's ok: https://drive.google...?usp=share_link

 

Update: if GIF files are not starting animation directly from goolge drive link, please download them.

Not all of my devices starting the animations automatically.

Edited by andriy_melnykov, 04 February 2023 - 07:59 PM.

[UnityLover]

Wow! I can see it wobble slightly! Is that it's rotation? or just a little processing problem.

[TOMDEY]

Interesting! I've been trying to see it "real time" visually (strobed of course) using Night Vision and a Big telescope. I've been looking at it with the Gen 3 Night Vision on my 29-inch (and now 36-inch) telescope with a high-end Thorlabs optical chopper that I modified to be extremely precise and of course tunable. I note that every now and then the star flashes (seems to flash) a lot brighter than average. Aperture of course helps. I'm still not satisfied and want to clean up the performance. I am working on refurbishing the dome this coming spring... and will probably get back to it the following season. PS: The image has to go directly to the GaAs photocathode, so it must be chopped before it gets to that. The down-stream phosphor has a relative long decay time, so the strobe freq must be like 0.2 to 1 Hz to make it most obvious.    Tom

Attached Thumbnails

• 

[andriy_melnykov]

Wow! I can see it wobble slightly! Is that it's rotation? or just a little processing problem.

Hi!

Thanks :-)

The wobbling is of course the postprocessig effect. I think, 3 factors can contribute to it:

- atmosphere effects

- noise + filtering in postprocessing

- aligning of the animation frames

 

Also, contrast gain on the Crab Pulsar star is much higher, as for other stars, thats why the contribution of postprocessing errors is also higher

Edited by andriy_melnykov, 05 February 2023 - 05:43 AM.

[andriy_melnykov]

Interesting! I've been trying to see it "real time" visually (strobed of course) using Night Vision and a Big telescope. I've been looking at it with the Gen 3 Night Vision on my 29-inch (and now 36-inch) telescope with a high-end Thorlabs optical chopper that I modified to be extremely precise and of course tunable. I note that every now and then the star flashes (seems to flash) a lot brighter than average. Aperture of course helps. I'm still not satisfied and want to clean up the performance. I am working on refurbishing the dome this coming spring... and will probably get back to it the following season. PS: The image has to go directly to the GaAs photocathode, so it must be chopped before it gets to that. The down-stream phosphor has a relative long decay time, so the strobe freq must be like 0.2 to 1 Hz to make it most obvious.    Tom

Hi!

Wow, 36 inch 

I was also playing with thoughts to see the pulsar visually, also with a night vision device.
Do you use a 50% slots wheel, like in your photo? In my opinion, 50% is far from optimal for the Crab Pulsar. If we look at the brightness curve of the pulsar (wikipedia link): https://upload.wikim...if/440px-M1.gif

we can see two pulses during the period. 50% slot will do a "sliding filter" on a curve, and additionally will mostly only show the difference between the two pulses.
I was using 25% slots, see my photos. And I would not go higher than that, rather lower.
I would also go with 5-10 slots to reduce the rotation speed and minimize the possible vibrations.
Do you have a possibility to change the wheel in your stroboscope?
Alternatively you can try to use a double frequency for your wheel - to see the differences between sum of both pulses and pauses between them.

Edited by andriy_melnykov, 05 February 2023 - 05:41 AM.

[Mert]

Extremely well done, congratulations!
That is a very nice tecnical project, good job!

[Rasfahan]

Amazing project. I didn‘t even know about this approach to imaging/visual. Would you think, with better tracking and more aperture, it would be possible to do this without a stroboscope? I. e. to capture a longer video feed, bin the frames post-hoc according to the pulsar cycle and then stack those frames?

[TOMDEY]

Hi!

Wow, 36 inch 

I was also playing with thoughts to see the pulsar visually, also with a night vision device.
Do you use a 50% slots wheel, like in your photo? In my opinion, 50% is far from optimal for the Crab Pulsar. If we look at the brightness curve of the pulsar (wikipedia link): https://upload.wikim...if/440px-M1.gif

we can see two pulses during the period. 50% slot will do a "sliding filter" on a curve, and additionally will mostly only show the difference between the two pulses.
I was using 25% slots, see my photos. And I would not go higher than that, rather lower.
I would also go with 5-10 slots to reduce the rotation speed and minimize the possible vibrations.
Do you have a possibility to change the wheel in your stroboscope?
Alternatively you can try to use a double frequency for your wheel - to see the differences between sum of both pulses and pauses between them.

Yes, yes, and yes. If you look at my wheel there... I modified it by placing the high-freq source/sensor on the outer edge and the low freq chopping inboard. I can vary the duty-cycle and relative phase of the inner two apertures by rotating those blades around relative to each other. And the Thor also has a very convenient phase adjust at the control box. I tested the phase-lock in the house by having the chopper looking at a (convenient!) 30 Hz LED running off the regulated line. Varying the intensity of the LED varies the pulse-width. Then I ate lunch and returned --- finding that the tuned Thor (running on its own internal clock) was still dutifully locked in phase on the line feed (to the source). And... because I'm only looking for the beat modulation "real time", it doesn't have to be all that precise to at least detect the pulsing of the star. If I want to actually measure that freq well enough to do some actual ~science~ --- I would pretty much have to run the thing continuously and uninterrupted --- for most of the night? And then maybe try to detect the phase changes as e.g. the (ostensible) plate tectonics on the surface of the Neutron Star to lead or lag the hot spots from their long-term averages. I doubt that I'll ever get to that. Just want to comfortably ~see~ it strobing well enough so any friend can just ascend the ladder, look into the (Night Vision) eyepiece, and declare, "It's Blinking!" OH! The star that just happens to be right next to it provides a superb photometric reference... so that even smaller variations in the pulsar can be detected by taking the running ratio of their intensities. This substantially normalizes out atmospheric effects. This also helps visually.    Tom

 

Page from my very early notes when I was first examining it with the 29-inch scope under pretty crummy conditions. It was bitterly cold than night! >>>    Tom

Attached Thumbnails

• 

[james7ca]

Nicely done and the animation seems quite convincing.

[andriy_melnykov]

Extremely well done, congratulations!
That is a very nice tecnical project, good job!

Thanks!

 

Nicely done and the animation seems quite convincing.

Thanks!

 

Amazing project. I didn‘t even know about this approach to imaging/visual. Would you think, with better tracking and more aperture, it would be possible to do this without a stroboscope? I. e. to capture a longer video feed, bin the frames post-hoc according to the pulsar cycle and then stack those frames?

I have also thought this way first. Even did some captures with my gear, but then did some calculations. Calculations show, this is not possible without photon counting camera like EMCCD or iCCD or similar. Basicaly I looked, how many photons can I get from the pulsar per pixel with short captures, like 8 ms (25% of 30Hz period). If it is below like 3-5 photons - not possible, because of read noise. The problem is (according to my understanding, please correct me, if I'm wrong), if our signal is under read noise, don't matter, how long we stack, signal will be always under noise.

But one can caculate, for example, what minimum aperture is needed to realize this approach. I did calculations only for my telescope.

 

Update: I was wrong, see post #16 below.

Edited by Dan Crowson, 06 February 2023 - 01:06 PM.

[andriy_melnykov]

Yes, yes, and yes. If you look at my wheel there... I modified it by placing the high-freq source/sensor on the outer edge and the low freq chopping inboard. I can vary the duty-cycle and relative phase of the inner two apertures by rotating those blades around relative to each other. And the Thor also has a very convenient phase adjust at the control box. I tested the phase-lock in the house by having the chopper looking at a (convenient!) 30 Hz LED running off the regulated line. Varying the intensity of the LED varies the pulse-width. Then I ate lunch and returned --- finding that the tuned Thor (running on its own internal clock) was still dutifully locked in phase on the line feed (to the source). And... because I'm only looking for the beat modulation "real time", it doesn't have to be all that precise to at least detect the pulsing of the star. If I want to actually measure that freq well enough to do some actual ~science~ --- I would pretty much have to run the thing continuously and uninterrupted --- for most of the night? And then maybe try to detect the phase changes as e.g. the (ostensible) plate tectonics on the surface of the Neutron Star to lead or lag the hot spots from their long-term averages. I doubt that I'll ever get to that. Just want to comfortably ~see~ it strobing well enough so any friend can just ascend the ladder, look into the (Night Vision) eyepiece, and declare, "It's Blinking!" OH! The star that just happens to be right next to it provides a superb photometric reference... so that even smaller variations in the pulsar can be detected by taking the running ratio of their intensities. This substantially normalizes out atmospheric effects. This also helps visually.    Tom

 

Page from my very early notes when I was first examining it with the 29-inch scope under pretty crummy conditions. It was bitterly cold than night! >>>    Tom

Ok, now I understand your gear better, thanks.

These are interesting plans!

When you get it working, would you also take a video of the pulsar for us, please?   Maybe will be better visible with an eye, but still, it is interesting.

[Rasfahan]

Thanks!

 

Thanks!

 

I have also thought this way first. Even did some captures with my gear, but then did some calculations. Calculations show, this is not possible without photon counting camera like EMCCD or iCCD or similar. Basicaly I looked, how many photons can I get from the pulsar per pixel with short captures, like 8 ms (25% of 30Hz period). If it is below like 3-5 photons - not possible, because of read noise. The problem is (according to my understanding, please correct me, if I'm wrong), if our signal is under read noise, don't matter, how long we stack, signal will be always under noise.

But one can caculate, for example, what minimum aperture is needed to realize this approach. I did calculations only for my telescope.

I think if you stack enough, read noise will only increase with the square root of the number of exposures, while signal will increase linearly, and the other noise terms add in quadrature. So you’ll probably need a long video which means lots of data. On the other hand, the FoV is small, so with good tracking/guiding it might be possible to acquire, say, a 1h (4h?) video, do temporal binning in post and stack for each bin.

 

But I just did a back-of-the-envelope estimate and I get to approx. 10-100 photons/s (depending a lot on aperture, and the exact spectrum). So if we do 120Hz video we‘re having about 0.1-1e- signal per frame on average (assuming unity gain - that‘s a good camera we‘re having here). Assuming we can bin that perfectly (so 4 „bins“ per cycle at 30Hz), that’ld mean we’ll need about 1000s for a SNR of 17-30. Add in shot noise (I’m at Bortle 5… that’s a lot of photons) and quantization noise from the binning and it adds up. And we‘re still discounting any variance in the framerate. Getting a totally stable stream of 8ms exposures isn‘t easy.

 

So yes, your approach is certainly superior.

Edited by Rasfahan, 05 February 2023 - 03:22 PM.

[freestar8n]

Amazing project. I didn‘t even know about this approach to imaging/visual. Would you think, with better tracking and more aperture, it would be possible to do this without a stroboscope? I. e. to capture a longer video feed, bin the frames post-hoc according to the pulsar cycle and then stack those frames?

I know someone who tried this approach with I think a 14" sct, but didn't have adequate SNR for it to work.  But that was some years ago before cmos video with low read noise and high QE - so it may be feasible now.  With enough aperture I think it should work.  Some cameras can gps timestamp each frame - but as long as the frame rate is constant and there are no dropped frames you should be able to stack the frames in the right timeslot.

 

Yes - there is no need for the signal to be above read noise in each exposure.  Stacking always causes the object signal to rise above sensor noise as long as pattern noise has been calibrated away.  Constant dithering would be good also.

 

That's a nice result by the OP - congrats.  And welcome to cloudynights.

Frank

[Robert7980]

That's excellent work!

[andriy_melnykov]

I think if you stack enough, read noise will only increase with the square root of the number of exposures, while signal will increase linearly, and the other noise terms add in quadrature. So you’ll probably need a long video which means lots of data. On the other hand, the FoV is small, so with good tracking/guiding it might be possible to acquire, say, a 1h (4h?) video, do temporal binning in post and stack for each bin.

 

 

I know someone who tried this approach with I think a 14" sct, but didn't have adequate SNR for it to work.  But that was some years ago before cmos video with low read noise and high QE - so it may be feasible now.  With enough aperture I think it should work.  Some cameras can gps timestamp each frame - but as long as the frame rate is constant and there are no dropped frames you should be able to stack the frames in the right timeslot.

 

Yes - there is no need for the signal to be above read noise in each exposure.  Stacking always causes the object signal to rise above sensor noise as long as pattern noise has been calibrated away.  Constant dithering would be good also.

Sorry, I was wrong here regarding signal under read noise! Looked again at SNR formulas - it is of course possible to get signal after stacking, if subs have a signal, smaller than read noise.

Actually I have tried it with my scope and didn't get any stars at all with many 8 ms subs. But I probably did something wrong. Maybe will try it one more time.

Edited by andriy_melnykov, 06 February 2023 - 12:37 PM.

[Rasfahan]

Sorry, I was wrong here regarding signal under read noise! Looked again at SNR formulas - it is of course possible to get signal after stacking, if subs have a signal, smaller than read noise.

Actually I have tried it with my scope and didn't get any stars at all with many 8 ms subs. But I probably did something wrong. Maybe will try it one more time.

It'll probably be very difficult - my estimate above was for 300mm aperture, discounting camera QE, losses in the optics and, of course, sky glow and LP - I think it's quite an uphill battle. If the clouds here ever clear long enough to get another shot at M1, I'll try, too.

[CliveFX]

This is all so impressive. I was thinking of doing this with a lock in amplifier and a single large photosensor.. looking at it visually is such a cool idea

[freestar8n]

It'll probably be very difficult - my estimate above was for 300mm aperture, discounting camera QE, losses in the optics and, of course, sky glow and LP - I think it's quite an uphill battle. If the clouds here ever clear long enough to get another shot at M1, I'll try, too.

In terms of doing it without a rotating shutter and just using video, then binning exposures with similar phase in the lightcurve, there is no difference in the factors you list compared to using a shutter.  The only increased noise term is read noise since all the exposures are short.

 

But at the same time you are always exposing and collecting light 100% of the time, as opposed to the shutter approach that spends much of each exposure with no light coming through.  So it's possible you would have higher SNR in a given time - as long as read noise is small.

 

So the main things needed are a video camera with high QE and low read noise - combined with large aperture.  A key requirement on exposure time and sensitivity is that you need at least one star visible in each exposure so you can align and stack properly.  You don't need to see the actual pulsar, though.  It should emerge in the stacks.

 

If you wanted about 8 samples through the full period that would mean about 4ms exposures.  No need to worry about saturation so just crank the gain on a modern cmos camera to make the read noise as small as possible.  If you can see a star in view you know it is at least possible.  If you don't see a star it may still be possible as long as you guide very tightly - but you won't be able to dither since you can't align the stacks without a star.

 

Frank

[smiller]

Well done!   I love citizen science projects!

[andriy_melnykov]

This is all so impressive. I was thinking of doing this with a lock in amplifier and a single large photosensor.. looking at it visually is such a cool idea

Hi, can you please explain it a little bit? I know, what is lock-in amplifier, but what would you use as a photosensor and optic? One telescope for photosensor and lock-in device and another one for camera for normal capture?

[andriy_melnykov]

Hi all,
after posting this topic with my work for Crab Pulsar from 2022, I was thinking, if I can do it better?
I was also wondering, why do I need so complex postprocessing, if I see the pulsar already after 5 minutes of integration time?
And unfortunately, I have found a mistake! For calculation of pulsar frequency, I didn't consider a Barycentric velocity correction! Pulsar frequency in the catalog is defined relative to barycenter, frequency seen at our Earth is different because of doppler shift. Recalculation has shown that I shouldn’t see any pulsation at all, stroboscopic period was completely wrong (like 100 seconds instead of 80 seconds).
So, honestly to say, I don’t know what pulsation I saw in the data from 2022 - maybe it was some harmonics of stroboscopic frequency, maybe only noise, and I probably tricked myself.
I can try to use data from 2022 again with correct calculation (and I want to do it later), but I decided to do it in another way.

[andriy_melnykov]

I think the stroboscope mode of an optical chopper with “sliding window” is good for visual observation, but not for imaging. It would be better to have a stable phase relation between the chopper wheel and pulsar, and a possibility to change the phase on demand.
Of course it is an ambitious goal to have stable phase relation for many hours without phase locking, but should be doable with GPS synchronization. Actually, the stability requirement is the same, as for “sliding window” mode in case of imaging.
So, I updated my scripts to calculate the frequency of the pulsar with barycentric correction (used this site for velocity correction: https://astroutils.a...t/barycorr.html ) and updated stroboscope software to be able to define the wheel speed with very high resolution - frequency can exactly equal the frequency of the pulsar. I also added a button that produces a phase shift addition of 1/8 of the pulsar period.
The idea of the session is to take subs for one animation frame, then press the button for the phase shift, take subs for another frame and so on, 8 frames for the whole period, each frame with a stable phase. Also, more than 8 frames should show the repetition of the pulsar pattern.
So, I tried it one night, and it worked flawlessly! I do control imaging sessions with parallel EAA, to control the quality of the subs - and was very excited to see the pattern of the pulsar already online! Also, after 8 frames the pattern repeats exactly. I am pretty sure that the system is working correctly now.
I took 10 minutes of 8 second subs (limitation of my mount) for each animation frame. Gear was the same, as in 2022, except I have used a 0.63 reducer. I think the reducer only made it worse (stars FWHM roughly stays the same for my system, but the brightness of the Crab Nebula in background should be higher).
As postprocessing I did only the stacking of each frame in DSS, stretching and some noise reduction in Photoshop. I also stacked intermediate frames from mixed subs of adjacent frames, to make animation more smooth.
Here is the animation with 24 frames: https://drive.google...?usp=share_link
And I also added the animation with only 4 frames directly to the post (because of 500kB limit).
The pulsar pattern is clearly visible, also the difference between the first and the second pulsar pulses is visible.

 

 

Attached Thumbnails

•