interesting... I was finding that numbers less than 2000 were just not moving the star very far, but perhaps I'm thinking the wrong with when it comes to PHD.
With CCDSoft (what I'm used to) your aim is in the one movement to move the star a good distance across the screen such that CCDSoft can re-analise the whole FOV, detect where the star has moved to and then have a good number of pixels in movement to get a reasonable vector. I'm now realising the contrast between that and PHD's method where it follows the star bit by bit for x steps.
I'll see if I can get that majical 30 or so steps
.. I think that will be a value of about 1500 for me.
Thanks,
Roger.