tripgood I used bin 2 last night and was able to guide with a max of 3 stars by pushing the gain past 40db.
This (40 dB gain needed) was also what I was seeing before the Build 29 Beta, with a 55mm/f4.55 guide scope, same camera and filter.
Let me explain what could be happening with all this mess:
(1) we know that saturated stars cannot be used for guiding. The centroids cannot be located with any kind of accuracy, and we need to be able to locate the centroid to 1/10 of a pixel.
(2) we also know that that stars that are too dim will also produce poor guiding cause by lower signal-to-noise ratio from both sky background and camera noise.
(1) and (2) together means that you need to have good dynamic range to get enough stars for milti-centroid guiding.
From what I was told, one of the things that changed from v1.8 was using a untrucated sum when performing the 2x2 binning. I.e., the data becomes a 10 bit value. If the least significant bit of the sum is maintained, we actually achieve a 3 dB improvement from binning using simple statistics (variances add, so standard deviation improves by square root of it). This was what v1.8 used.
In the early v1.9 betas (as I was later told), ZWO had switched to using an 8 bit value after summing (i.e., 10 bit number shifted down by 2 bits and trucated, into an 8 bit integer again). Notice that quantization noise now predominates the SNR term.
When you look at the distribution of star intensities, you will find there are more stars at magnitude 8 than magnitude 7. More stars at magnitude 9 than at magnitude 8, etc. So, to get lots of stars to guide with, the deeper you go the more stars you will have. But as you go deeper, the SNR gets worse.
As you change the camera gain, you are basically moving the range of guideable stars uo and down the magnitude scale. For a fixed exposure time, when decrease the gain, and you are picking only the brighter stars -- that is good for SNR, but you have fewer stars.
When you increase the gain, you are seeing more of the lower magintude stars, but your SNR suffers.
So there is a "goldilocks" region (sorry -- using the term we used during my involvement with the NASA SETI program back in the 1980s :-) for the gain setting.
This is where a higher dynamic range (10 bits) helps tremendously. You can use lower gain, and still see dimmer stars above the quantization noise. This is what I noticed with the Build 29 beta. I could dial the gain back below 30 dB and still find stars (11 to 12 usually).
The next thing to look at is the star size. If the star size is much less than one pixel, it is also impossible to get an accurate estimate of the centroid. Now, a star size of 0.5 pixel does not mean that the star is completely inside a 0.5x0.5 pixel square. Only half (that is the "H" of HFD) of the star intensity is inside the square. The rest is spread farther than 1/2 a pixel.
One of the bugs that was introduce in 2x2 binning was that ZWO had kept the same star size limit as they used by 1x1 binning. This of couse is a poor choice (I had alluded to it in my very first report of v1.9 guiding). A "pixel" in 2x2 binning is now twide the size of the original 1x1 pixel. From a March 16 email, I was told that "we also half the select box of star." I think that means that whatever they used as 1x1 binning, they halved it for 2x2 binning. So it won't be rejecting as many stars.
The HFD of a star will depend on the the aperture (determines the Dawes/Airy limits of optical star size -- it is not infinitelsimally small, but limited by the Fourier Transform of the aperture), the focal length (determines the optical star size in pixels) and just as important -- "seeing" (i.e., atmospheric turbulence).
Roughly speaking, the star size that you see on the sensor is a convolution of the Airy disk of the star with the probability density of the turbulence. At sea level, and good seeing, the turbulence is about 2 arc-seconds. Since the turbulence is statistically independent from the Airy disk, the convolution result follows the sum of variances. I.e., the star size as seen by the sensor is now sqrt( A2 + T2) where A is the Airy disk, and T is the turbulence, both in arc-seconds. Use the plate scale (i.e., focal length and camera pixel size), and you have the star size in pixels.
So, it is indeed possible that perfect seeing can cause ASIAIR to pick fewer(!) stars, because all the stars are now below the star size limit! If you are on top of Muana Kea, binx2 may not pick a single star because they are all too small :-).
Now, all this really only matters with the combination of the ASIAIR and mount. We both use a RST-135, and guiding rate has to be very high. The problem is that with the ASIAIR, we have to use binx2 to acchie this frame rate. A better processor will be able to achieve 2 FPS with 1x1 binning, and the "goldilocks" region is much wider.
From what I can see from their Periodic Error curve, I suspect that ZWO's mount may need as high a guiding rate to produce better than 0.35" total RMS error that we get from the RST-135. So it behooves them to make guiding work well on the RST-135; otherwise, the ASIAIR will also not be able to guide their own mount well. The first derivative of the periodic error curve can be very large and you need very fast feedback to keep up, otherwise the stars in each frame will be a short streak instead of a point. I suspect that autoguiding with adaptive optics will work well with something like a harmonic drive gear.
Chen
