c131frdave Hoping Chen sees this post and explains it in layman's terms.
I'll try to get as layman as I could, but this is a technical hobby (the Facebook selfie taking "astrophotographers" not withstanding).
First thingto consider is that there is a pretty big difference between the classical CCD in-situ binning versus the CMOS off-sensor binning (since the commercial CMOS sensors don't support in-situ binnng),
Lets make things simple and assume there are only two camera noise sources, the read noise and the dark current noise (you will see why I pull these two out among a myriad of camera noise, like the CMOS fixed pattern noise, etc).
Let R be the read noise power and D be the dark current noise power.
Let S be the signal (thus S2 is the signal power).
The signal-to-noise ratio for a single pixel is therefore S2/(R+D). The reason you can add the denominator is from assuming that the read noise and the dark current noise are statistically independent -- i.e., "variances add."
Now, lets now assume that you are a little oversampled, so that two adjacent pixels receives the same signal power. In the classic CCD, when you bin two adjacent pixels together (e.g., 2x1 binning), you only have to do a single readout to get the (Sx2)2 signal. While the dark current noise still sum.
What we see in the cassic CCD case is that the SNR is now more like (S+S)2/(R + 2D), or 4.S2/(R + 2D).
Notice that is R predominates over D (i.e., R>>D) then the original SNR is S2/R and after the 2x1 binning, the SNR is 4.S2/R. 6 dB improvement.
However, if read noise is much smaller than dark current noise, then the single pixel case gives a SNR of S2/D while the 2x1 binned SNR is 4.S2/2D, of a SNR improvement of "only" 3 dB!
(You probably see where I am going if you recall that dark current is proportional to exposure time, while read noise does not ... :-)
OK, now on to the cheap CMOS case...
With CMOS cameras, there is no way to read the 2 side-by-side pixels with a single read operation. If you want to bin CMOS, you read the two pixels and add them externally. The adding can be done in the camera's processor, or at the computer (the latter requires more data to be downloaded).
So, in this case, the single pixel SNR is still S2/(R+D) like the CCD case, but a 2x1 binned SNR is now (S+S)2/(2R + 2D), or 2.S2/(R+D).
Notice that whether R >> D or D >> R, we get a constant 3 dB of improvement.
The moral of the story is that with CCD, you are only going to get the 6 dB SNR improvement for short exposures (i.e., when R >> D). For long exposure CCD, and CMOS binning, you get 3 dB per doubling of pixels.
Still, 3 dB is nothing to sneeze at (a 2x2 bin can get you up to 6 dB SNR improvement). (Especially after ZWO woke up in v1.9 and decided to sum the 4 pixels into a 10 bit number instead of scaling the sum of four 8 bit numbers back to an 8-bit number).
In addition the SNR improvement, binning can improve (or worsen) the plate scale. If your guide star size at bin1 is about 3 pixels or larger, bin2 in ASIAIR should help. If bin1 star size is smaller than 2, bin2 will hurt since the star mass will be completely inside a 2x2 superpixel and there is no way to see any centroid change until the star moves across a superpixel bundary.
There is a further side effect in ASIAIR with bin2. Because of the way ZWO implemented the guide implementation, with bin1, you can't get anything faster than 1 frame per second, even if you reduce the exposure time to 0.25 seconds. However, you can get 2 FPS (important with mounts like the Avalon M series and RainbowAstro) at bin2; even if you use a larger sensor like an ASI178. The frame rate limitation, applies only to the ASIAIR (I get 9 FPS for the ASI290 with other software running on the same Raspberry Pi 4).
By the way, a gain of 60 (6 dB) is quite low (although the ASI290 seems to have already switched itself to high-conversion gain mode at that gain). You might want to try gains of 200 (20 dB) or even more, to achieve a lower read noise. At short exposures, you ware not limited by dark current noise; so, reducing the read noise might be beneficial.
For autoguiding, Bin2 is not for everyone, but can be useful if yor instruments fall into that zone. Your oversampled case seems to be able to take advantage of it, without having to switch to a camera with larger pixels.
Chen
