There's no magic. Higher gain means BOTH more signal AND more noise in each sub. You can see more with fewer subs but a clean final image still takes just as many hours with Gain=100 as with Gain=0.
My own preference is Gain=0 for "traditional" DSO imaging (many hours to many nights per final image), because I get better dynamic range in each sub.
OTOH, I juice up the gain as high as tolerable (200? 300?) for live video (EAA), so I can see a lovely color image quickly, and just put up with the extra noise.
Higher gain means less noise per sub. You are making a common mistake by looking at the data "the wrong way", in ADU form rather than electron form. In electron terms, higher gains have less noise, and this can be demonstrated through testing. Here is some test data, represented in different ways, from an ASI183:
The top row here is what most people see if they simply stretch a raw dark or bias frame (or even a light frame) from a camera, and "see" that at higher gains, there is "more noise". Problem is, your eyes deceive you! There is NOT actually more noise...the higher gain means that each electron is represented by MORE ADU, which makes it SEEM like there is more noise, simply because AS REPRESENTED, there is a greater deviation around the mean. But this is a RELATIVE thing, not an absolute thing.
The bottom row here is after the data was re-converted back to an electron representation. In this representation, a representation of the true analog signal, the drop in read noise becomes quite apparent...and the higher gains are exceptionally clean. Not just from a spatially random noise standpoint, but also from a pattern noise standpoint.
This is the image form of the data. You can also take any slice of this data and render a 2D noise plot from it. Now, a slice through the middle of the top row results in something like this:
Again, it APPEARS as though higher gains have more noise. But again, this is simply because each electron (an absolute unit) is represented by more ADU (a RELATIVE unit) at higher gains than lower gains. In fact, this increased representation by ADU improves your ability to detect faint signals (small signals, few electrons), as with more ADU each electron, and all the noise within the signal, is more accurately represented. Note that while the lower gains with fewer ADU seem to have a smaller distribution...note the "stepped" or "blocky" nature of those plots...there are not all that many different possible points in the plot. This is quantization error.
This final example is a slice through the middle of the bottom row. The high quantization error at the lower gain is now quite apparent, while at higher gains the high precision and very low distribution is quite clear:
These plots demonstrate the true nature of what is really going on, before an arbitrary and relative unit, the ADU, obscures reality and misleads you into thinking high gain means more noise.
Now, you can "swamp" the read noise at any level, and effectively bury the differences between gains. This usually means long exposures at low gains, medium exposures at middle gains and short exposures at high gains (all on a relative scale of course...there is no absolute time here...short may be 10 minutes at high gain, if you are using 60 minutes at low gain...)
In this plot you can see how increasing the shot noise in each sub ultimately results in the signal, its noise and its distribution being totally dominated by what you get from the sky. Now, inherently, for a given exposure length, the amount of signal in a sub is the same...doesn't matter what gain. The signal you get from the sky is fixed for a given TIME. However, you can CHOOSE how long you expose for at any gain. You may clip at some gains, you may not swamp the read noise at others, but how much signal is in each sub is dependent on TIME, not gain. More time with the shutter open, the more signal you get. Increasing gain does NOT increase the amount of signal...it only reduces the amount of noise. So higher gains does not mean both more signal and more noise. On the contrary, usually, higher gains, since you also lose dynamic range with most cameras at higher gain, means less signal...and also less noise since read noise drops as well. Used properly (same total integrated time), you can get the same signal at a high gain as at a low gain. The differences is not necessarily how much signal, but how well the faintest signals are represented...and that is where higher gains can often do a better job with more reasonable exposure times (2, 5, 10 minutes, rather than 30, 40, 60, 90 minutes, per sub).
Edited by Jon Rista, 29 April 2020 - 07:15 PM.