To measure the effectiveness of reduction of centroid error in ASIAIR v1.6 Beta's multi-star autoguiding, I again used my mount simulator.

The simulator uses an ASI174-mini as a guide camera, with an equivalent focal length of 212mm; so a moderately short "guide scope." (I usually use 200mm and 250mm guide scopes for my 454mm focal length OTA.)

I added Gaussian noise to the x and y axis of each star (every star has its own independent noise values) in the gnomonic tangent plane that is centered at the RA and declination of the sky. Since the x and y are independent Gaussian noise, the centroid should have a Joint Gaussian probability distribution, which is radially symmetric; what you would expect star movement due to atmospheric turbulence to look like -- i.e., no bias in any direction.

For the image captures, I used the Pleiades as the star field because there are lots of bright stars (my simulator currently uses the Hipparcos catalog which only goes down to 9th Magnitude stars -- I will soon be adding stars from the THCHO 2 catalog).

I also turned the MAX RA and Declination pulses to 1 millisecond (the minimum) and turned RA and declination aggressiveness to their minimums (5%). Together, these causes practically no corrections to the mount, so the guide graph should show the plot of the centroid of the guide star (or the average centroid, in the multi star case).

This is the guide graph for the familiar single star case:

Without changing anything else, I switch to multi-star autoguiding (i.e., allowing ASIAIR to pick its own guide stars):

Since there is practically no guide pulses (1 millisecond), the guide graphs and RMS errors should merely show the changes to the centroid of a star (or the average centroid of multiple stars).

Notice that the centroid estimation RMS error dropped from 1.68" to 0.70". If, in the past, your guide error is limited by "seeing" (and not by mount mechanicals), you can expect the same kind of improvement (a little better than a factor of two). If you were limited by mount mechanics, then multi-star will do nothing for you. I suspect that most people should fall somewhere in between, and over time, perhaps ZWO can improve the algorithm (with 12 stars, they should be able eek out a little more than a factor of two).

I did notice that the frame rate had dropped; perhaps processor bound? In the coming days, I will be trying a USB 3 based ASI290MM instead of the USB 2 based ASI174-mini above (which also has a larger pixel count).

Until I can test under real stars to know for sure, the multi-star guiding looks promising for now.

Chen

I live in Japan, and the latest version of TestFight is still 1.5.3.

    Hikaru I'm relieved, and I'll wait patiently.

    I have no idea why, Hikaru san. I do have an Apple Developer account, but I didn't think that makes any difference.

    Chen

      The "New Plate-Solving Algorithm" under "Experimental Feature" works quite well.

      I am running an FOV of 3.07º x 1.93º on my mount simulator, and by reducing the gain and exposure time of the ASI174-mini like crazy, I could finally get the "usual" plate solving to fail after 110 seconds (even with a perfect focal length). I then turned on the new plate solving algorithm and it solved in less than a second.

      Lets hope the new plate solve works also when polar aligning. Perhaps we can now do polar alignment even earlier than before.

      I like ZWO's new approach, which appears to be placing improved technical functions (which people depend 100% on) under "Experimental Feature." If there is some fatal flaw in any single function, you can just revert to the old scheme, instead of reverting the entire app and firmware and lose all new functions.

      Chen

      w7ay it is strange that no one else is getting the 1.6 after almost 2 days.

      Another piece of good news, there one more improvement to autoguiding: the default MinMo has changed from 0.2 pixels to 0.1 pixels. I check the PHD2 logs for the RST-135 and the AVX mounts, and both logs show MinMo of 0.1 pixels.

      Hysteresis remains at 0.1 pixel.

      Best to have both adjustable, but I will settle for this -- some of us have been pleading for it for over two years!

      I haven't tried to simulate other mount types to see if this is a global change for all mounts, or for all plate scales (guide scope focal lengths).

      People with short focal length (less than 300mm) guide scope, with large periodic error mounts that have low backlash should be happier with a smaller MinMo.

      Chen

        w7ay That is good news, too bad it still not adjustable hopefully I can get a better guiding out of my AVX mount.

        w7ay

        What settings are best for OAG users?

        And for multistar guiding option, why those with lower end mounts won't improve their guiding?

        Andrej

        • w7ay replied to this.

          astrosatch And for multistar guiding option, why those with lower end mounts won't improve their guiding?

          Using multiple stars essentially reduces the centroid measuring error.

          Atmospheric turbulence ("seeing") jitters the star's position as seen by your guide camera. When you use longer exposure times, you are averaging the centroid value, and if the turbulence causes symmetrical errors (things like upper atmospheric winds could cause a bias, though) then the star image is more bloated, but the centroid is more accurate. When you double the exposure time, the variance of this error should decrease by a factor of two (the RMS error decrease by a factor 1.4). If you quadruple the exposure time, the centroid estimation error should reduce by a factor of 4 and the RMS reduces by 2, etc, etc.

          The problem with increasing exposure time is that the control loop becomes less effective (and can become oscillatory). And that is why you have to keep the aggressiveness low. If seeing is not a problem, you should use as short an exposure time as you have processing power (sampling theory holds for digital control systems).

          What multiple star centroid computation does is to use more stars instead of more exposure time.

          The atmospheric turbulence that affects one star is hundreds and thousand of kilometers from one another, so have no relationship (in Probability Theory, this is called " statistically independent", and all sorts of theorems apply). And according to Probability theory, if the statistics is Ergodic, averaging with multiple stars is the same as averaging a single star over a longer period. I.e., with two stars, the variance of the error should reduce by 2 (RMS by 1.4), with 4 stars, the variance should drop by 4 (RMS by 2), etc.

          So, what multiple stars does is to reduce the measurement error of the centroid, without having to increase the exposure time, (as mentioned above, longer exposure time is a bad thing). No more, no less.

          If you have a mount that has bad gears, insufficient motor torque, bad backlash, bad third axis balance, and if those errors are larger than the centroid error in the past, then multiple stars will not help (or help only partially).

          Measuring the centroid of multiple stars only helps reduce the centroid estimation error. It does not cure any mount mechanics.

          If your mount is erratic, a less noisy centroid might reduce problematic wild swings. But overall, there is not much you can do except to get a better mount -- the mount is often the weakest link in astrophotography, and where you should spend the largest portion of money on. If you pay more for your camera than for the mount, you should probably change dealers to get better recommendations.

          Using multiple stars should also reduce "star lost" cases, since seeing will not affect all the stars within the same exposure; the probability of losing all 12 stars at the same time is like winning a lottery. So, multiple stars will help a little with cameras with poor signal to noise ratio. I don't think it will help with real camera bugs though, like dropping out from USB or producing image tearing.

          Multiple Stars autoguiding is neither some magic, nor something new. It is merely a different way to reduce the centroid measurement error without having to increase exposure times. See this paper (DONUTS algorithm, 2013) in the professional world, for example:

          https://arxiv.org/abs/1304.2405

          Notice that the abstract mentions "high cadence," meaning short guide exposures.

          The DONUTS algorithm has been available in INDIGO for some time now (you should find the code in Github); before PHD2 implemented multiple stars. However, it requires very good signal to noise ratio -- the real astronomers just use bigger guide telescopes, and expensive cameras for guiding. INDIGO Sky runs on a Raspberry Pi.

          Chen

            w7ay Thank you Chen, this is really good stuff. Sadly, still no update as of now. I got clear sky tonight I want try it.

            • w7ay replied to this.

              Skylab1 Sadly, still no update as of now. I got clear sky tonight I want try it.

              Darn. I must somehow be on the pre-public Beta list then. Perhaps Sam did pull some strings -- when you are the chief bottle washer, you can make anything happen :-). Sorry for getting your adrenaline up, but at least you know what is upcoming; and probably soon too.

              It was only partially clear with 0ºC last night, so I did not go out (this is just a hobby, and I don't believe in suffering when there are other things to do :-). It should be better tonight, so I might have a chance to check out multi-star guiding with real stars. Hey, also to see if I can polar align when it is still dusk, using the improved plate solve.

              With Portland's unpredictable weather, you will probably be using v1.6 Beta before I have the chance to :-).

              Chen

                w7ay

                w7ay that ok I can just drool at the moment, but can’t wait to see your live data if you get a change to test it.

                w7ay just wondering is there a way you can ask when will the public beta be available for the rest of us?

                OK, a couple of guide graphs for you to drool over.

                Mount is an RST-135. Payload includes FSQ-85 with its 1.01x flattener, a ZWO M54 filter drawer and ASI2600MC. On a shared dual saddle plate is a Borg 55FL as guide scope and the USB 3 (pancake) version of the ASI290MM.

                Wind is gusty, peaking about 15mph.

                And a little later:

                Back in the ASIAIR v1.3 days, I was getting about 0.7" total RMS error. Somewhere between v1.3 and v1.5, the error had doubled to around 1.3" total RMS. The 0.5" is close to the best I have ever gotten outdoors. I have gotten 0.10" with my simulator (no wind, no atmospheric turbulence, no sky background, etc).

                Caveat: I was getting constant Star Lost when I started testing. I noticed that the stars are all still there, the ASIAIR just couldn't find them, perhaps after a gust of wind. I then defocused the guide scope so that the star size degraded to the 2.5 to 3.5 region. It is never a good idea to defocus anything because of the loss of signal to noise ratio. But it apparently worked to mitigate whatever bug is in the guiding algorithm. (I have already reported this by direct email.)

                Chen