Sonixx I understand oscillating needs to be avoided, but constant under correction is also not good.
Or am I missing a point here?
In one of the posts, I had mentioned that the the feedback mechanism in PHD2 violates the Routh-Hurwitz stability criteria if the loop gain (which is ad-hoc called "aggressiveness" -- you will never see that made up word in any Control Theory textbooks) is or over 1.0 (which is given as "100%" in PHD2) or if the control parameter (the measured centroid) causes the equivalent loop gain to be over 1.
The guide frames that are taken with mounts that have high periodic error slopes have stars that move in the frame. Guide stars are not stationary (they are never stationary, but the guide stars move correspondingly slower with mounts with slow slopes -- this is the sawtooth diagram -- once you grok the sawtooth, you will understand autoguiding loops).
Turns out it is possible to create a feedback algorithm that takes this centroid motion into account. But unless you are willing to write your own code, you will need to treat mounts with large slopes with extra care when using existing guide algorithms.
One of these things that you need to be careful of is the so called "aggressiveness" value, so that any centroid estimation error does not conflate with the mount's own error.
Do not try to correct for the centroid error in the feedback -- it, like wind gusts, will correct itself. Limit the corrections (by applying max pulse duration) to only allow as much correction as the mount's worst case error.
Look at your guide graph with correction pulses turned on. If you don't see a long constant comb of guide pulses all in the same direction, then your loop gain is more than sufficient, and therefore there is no reason to increase it (i.e., no need to use any extra "aggressiveness").
One other thing that you can do is to reduce the guide rate to lower than 0.5x sidereal. This again is from looking at the sawtooth. Note that the amplitude of the sawtooth is basically the guide error. And by reducing the guide rate, you also reduce the sawtooth's amplitude (the sawtooth amplitude is reduced too by decreasing the guide exposure time -- that is why it is imperative to use 2 FPS [0.5 second guide exposure]).
The RainbowAstro protocol allows you to tune the guide rate to a resolution of 0.01x sidereal rate. However, if you are using ASIAIR, you only have the choices of 0.25x, 0.5x, 0.75x and 0.9x sidereal rate (the ASIAIR is designed by simpletons for use by simpletons). If your mount's error is small enough to use 0.25x, choose that. If it is not small enough, use something like 0.3x or 0.4x sidereal rate -- fall back to 0.5x on ASIAIR since you don't have a choice. Do not ever use guide rates of 0.75x or 0.9x sidereal rate, since you are just allowing the sawtooth amplitude to be high, and I do not know of a RainbowAstro mount that requires that much correction within 0.5 seconds.
If your RST-300 mount's characteristics is similar to my two RST-135 mounts, you can get away with a guide rate of 0.25x sidereal (but the PE slopes are too large to work with 0.15x sidereal). And even with that slow 0.25x sidereal guide rate, I can set still the RA and declination pulses to be no more than 150 to 180 milliseconds each. Those durations are enough to correct for the periodic error slopes of my RST-135 gears. Since RainbowAstro guide pulses are sequential, be sure that N+M is less than 0.5 seconds when you are using 0.5 second exposure, where N and M are the max pulse durations. If that property is not maintained, there will be total corrections that take longer than your exposure time and the latency of an entire extra guide exposure occurs (i.e., an extra 0.5 second worth of uncorrected tracking).
Chen