Hi Matthew,
I only spotted it on your graphs after I had posted the other night but it appears that
you only sampled stars exclusively on the east side of the mount and therefore
missed sampling any with positive hour angles.
I cannot emphasize enough the merit in sampling stars across the whole sky,
which in the case of a GEM, includes performing a meridian flip.
It may be that your western skies are obscured or that for some reason you
can't perform a meridian flip, but if you can, I highly recommend on your next
sampling run you do so.
Some mount errors reverse their direction when the mount is flipped and this
signature can further help betray them. When you sample stars
on one side of the mount and then do the flip and begin to sample stars on the
other, don't be surprised to see your raw RMS pointing error residual suddenly jump up.
Quote:
|
Originally Posted by g__day
Thinking over the changes last night - it struck me as unusual to the lay person how whilst adding coefficients that decrease sigmas and hone RMS (by adding about 8 harmonic terms RMS went down around 18 - so +/- 9 arc seconds pointing - which seems superb - the polar alignment corrections swung by quite a bit - far more indeed than the improvement in RMS pointing - is that usual?
|
What your results so far have demonstrated is that polar alignment is only
one source of error within your system. When you consider how all the
other fabrication errors are entangled with polar misalignment in such a complex
and knotty way and when you consider the relative magnitude of the other
errors compared to the polar misalignment errors, then it will come as no
surprise that as you add or remove terms that are correlated with the polar
misalignment terms, that the polar misalignment terms will swing quite a lot.
Let us put it another way. If one were to try and determine the polar misalignment
without taking into account these other various sources of errors that have relatively
appreciable magnitudes, would one expect to arrive at a reasonable value for the
polar misalignment terms? The answer is, of course not.
Consider a simple exercise in statistics -
One wishes to determine the average age of all students attending a high school which
has a population of 1000 students.
One interviews the first two students and determines their average age to
be 14 and 14.5 and therefore calculates their average age to be 14.25 and the
sample standard deviation to be 0.35. Does this result likely provide a reliable
estimate of the average age of the entire population?
The more data you have, which means in this case the more of the sky you sample,
if the mount is free of random errors, then the more confident one grows of the
result when pointing to some arbitrary position of the sky.
There is no one single magic point in the sky to which one can align the polar axis
which then results over an extended period in zero field rotation for all RA/Dec points
in the sky.
In other words, if you do a drift test and tweak the mount to achieve minimal field
rotation at some point in the sky, don't expect that tweak to be as good across
the whole sky. In that sense, a drift test is only providing a small sample. If that
is the only point of the sky you plan on imaging and your imaging times
are sufficiently short that refraction and the other errors don't start to create
significant field rotation, then that may be all you need.
But if you want to achieve the best whole sky pointing performance and with
it the best chance of nailing where the pole is as a reference point, then the optimal
prescription is eliminating random errors with the mount/OTA, sampling a
large number of points across the whole sky, performing a statistical analysis of
the type TPOINT performs, optionally tweaking the polar elevation to help counter
for the effects of refraction in the area of the sky you are going to image and most
importantly, have a pointing kernel which is applying the same pointing model and
therefore the correct associated dynamic tracking rates.
Again I will be touching on some of this at Lostock.
By the way, an RMS of 18 arc seconds refers to a radius, so if the errors are
normally distributed, roughly 68% will fall in a radius of 18 arc seconds, which is
of course a diameter of 36 arc seconds.
Ernie mentioned using FO which as he points out is legitimate to do on a GEM
in a "if the glove fits" kind of way. However, because the TPOINT methodology
is to try and provide terms that mimic some expected mechanical phenomena,
then I would recommend that if it does not significantly improve the model
compared to those terms that have a ready and straight forward mechanical
explanation, such as HCES, then I would drop FO in favour of these more
less surprising terms. In other words, apply Occam's Razor. Then only use higher
order harmonics or polynomials or terms such as FO to 'mop up' anything that can't
be readily accounted for.
).
