[g__day (Matthew)]

I've switched from PHD to MaximDL back to PHD so I have a few comments.

First I image at long focal lengths (2300mm) for say 10, 20 or 30 minute shots maximum. I use a Lumicon OAG into a Meade DSI II Pro Mono 2*2 binned. So by my calculations each pixel in PHD is 2.2 arc seconds and in my main DSLR each pixel is 0.7 arc seconds.

Yesterday I shot M16 and combined just over 3 hours of shots (7 shots) - in DeepSkyStacker and was interested in the stacking error to give me a sense of the guiding precision between frames. The errors in pixels (dx, dy) from the first shot were (-0.01, -0.30), (-0.76, 0.21), (0.15, 0.08), (-0.25, 0.19), (0.12, 0.18) and (0.18, -1.25). So read that back as I understand the stacking processes - over 3 hours using combinations of 10 to 30 minute shots the net tracking error was 0.35 arc seconds - which is pretty impressive!

I usually guide on 3 second shots, and I focused first the main camera (using a Bhatinov mask) then the OAG CCD (Bhatinov mask on Jupiter - works well)! I find that either MaximDL 4.5 or PHD 1.10.6 do an equivalent job judging by tracking error in the correction graphs. MaximDL is more sophisticated - PHD appears simpler to use.

With PHD (given you have a tight, bright but not over saturated star - so the centroid calculation is fine) - you still have to think through your controls carefully.

For instance how do you minimise seeing errors (think of this as a signal / noise challenge - balancing false positives (over-steering) vs false negative (under correcting) guide commands):

1. Take longer shots so each shot averages the seeing wobble (bloating stars slightly) or
2. Take shorter shots and lower the RA aggressiveness to the smallest value that will track well without over shooting (but takes longer to correct, minimising over corrections though).

You need to see how stars are jumping at your image scale to guess this one right.

PHD can see well down into sub pixel movements - reasonable case 0.10 pixels, best case 0.01 pixels accordng to Craig. But take my case where seeing pixel is 2.2 arc seconds (with only average seeing conditions). Do I really think trying to track with 0.22 or even 0.44 arc second precision is optimal and achievable - that may be over reaching. The higher I raise minimum distance before a pulse is sent - the more I will not send false positive guide commands causing yo-yoing due to seeing errors (but the more true errors I let in). You have to get a feel for this.

Next is Hystersis in RA - some nights I set the RA aggressiveness to 30%, hystersis to 40% and get great results. I think by the screen shots Craig did in a debug run 1.10.7 will have a RMS error tracking bar - this will help optimising setting PHD parameters. Note Craig often seems to debug PHD on very short focal length shots - so his tracking graph is a near perfect straight line - but not as good as mine < +/- 1 arc second traking error). If I set RA aggressiveness right down OSC index goes from about 0.45 down below 0.20 - but occassionally corrections get large - +/- 1.5 - 2 pixels. With it higher > 75% OSC Index is around .045 but errors tend to stay within +/- 1 pixel.

Guiding isn't too hard - guiding really well is very challenging. My guiding is now sub arc second (but I still think there is room for a bit more improvement). Having used PHD for 2 years now and watched it grow I am very fond of it. I love MaximDLs plug ins - Ijust haven't invested sufficient time to try and master it.

I wish all these guide programs had a really smart brain to analyse your set up and optimise their parameters. I am trying to discover how to optimise auto-guiding - but its more brute force then elegance at present.

As a suggestion - why not have a thread on how to optimise PHD - folk can try different parameter settings so we can get more data on to optimise the parameters.

As a hint do you turn on PHDs error longs and look at how it is deciding how and when to switch and how far to correct? Do you run PHD Analyser on the normal log to check for bias or over correction in your tracking runs?

From experience guiding programs require significant investment to tune to their full capabilites - but the results once you get there are impressive!

Matt

PS

Fred - both PHD and MaximDL having tracking graphs - I like MaximDL's better as there are seperate graphs for each axis, whereas PHD super-imposes the errors on the one graph. However PHDs error log is a grat idea - it shows you how PHDs brain is thinking at each step, whether it is thicking to switch direction (and why) so - so you can get alot of control over how it is performing and why.