PHD2 dithering algorithms put to the test: is it time for an improvement?

[Leo Shatz]

I've conducted an experiment to evaluate the behavior and patterns of PHD2's dithering algorithms using a script in the Voyager. For this test, each sequence consisted of 400 dithering operations with a maximum displacement setting of 3 pixels in Voyager, magnified by a 5x scale factor in PHD2, resulting in a maximum displacement of 15 pixels. To track the movements, I made a temporary modification in PHD2 to mark each new lock position with a bookmark.

The findings highlighted a few issues with both dithering methods:

1. Neither method confined the scatter points within a reasonable distance from the start, allowing for significant drift.

2. The "random" method occasionally placed points too close together, creating a pattern similar to Brownian motion. This lacked uniform distribution across x/y directions and did not form a cohesive pattern that could be optimal for reducing noise for stacked images.

3. Contrary to expectations, the "spiral" method strictly followed a square spiral path with a constant step size and no randomness. This could inadvertently introduce repetitive patterns into stacked images, but I cannot be certain if this is really the case without running more extensive and complex tests.

Overall, these observations suggest that there is certain room for enhancing PHD2's dithering algorithms. I welcome any thoughts or experiences you might have had with PHD2 dithering. Do you think we need a new dithering pattern?

Thanks,

Leo

[Bruce Waddington]

Based on 10+ years of PHD2 support, I would say this isn't something worth fooling with.  Brownian motion is random, by the way, so no problem there.  Your complaint about the "expected" behavior of spiral dither is somewhat amusing.  Do you seriously think that "square corners" have any effect at all for dithering? <g>  Have you read the user guide description for spiral dithering including the term "dither with fixed size amounts"?  So I don't really see why the current behavior is "contrary to expectations"  I'm curious, do you do a lot of imaging using PHD2 and find that somehow the current dithering machinery is causing problems in stacking?  If so, that would be very surprising, this is simply not an issue that gets raised.  But if you have that problem, we would like to see your raw images.  You have to understand the context here.  In real imaging sessions, lots of things contribute to small movements of the target on the sensor - residual tracking and guiding errors, auto-focus runs, start/stop/pause/resume of guiding, even periodic auto-centering.  Changing something that has worked well for the life of the product is not something we like to do unless there is compelling evidence for it - not just notions that come from gedanken experiments.  When we choose to do that, we usually have to add things rather than replacing them because we have no way to know how outright replacement will affect the thousands of PHD2 users in the world.  And when we add options, we have to then educate people on how to make these choices and support them when they make bad choices or foul things up.

If you have available time and want to contribute to the PHD2 project, it would be a good idea to start by participating on the support forum, helping users deal with real-world problems and gaining a first-hand perspective on what things matter to them.  We would certainly welcome that, there are very few people supporting a huge end-user base.

Regards,

Bruce

[Leo Shatz]

Thank you for your feedback. While I greatly respect the invaluable and free support you offer to the PHD2 community, my primary intention here is to make the most of the PHD2 as its user (besides the efforts of a few last months I've spent developing the solar/lunar/planetary guiding extension for PHD2). My biggest concern, as I've stated in the original post is the unlimited drift that both dithering methods are creating. Also, my own imaging time is very limited and I'd like to make the best possible results with limited data that I'm able to collect. I have to admit that in my practice, however, I don't usually see problems with the drift getting too large, since I often pause  and restart imaging sequence to adjust some of the session parameters, but that doesn't mean that problems with dithering don't exist. I wasn't critiquing the square pattern per se; rather, I was sharing an empirical observation and taking a note that PHD2 doesn't confine dither points within a set radius. As for the The Browning motion in the "random" method - it is possibly not so random, as it shows signs of unidirectional and very densely packed movements, although I haven't performed any statistical analysis of it.

Here is my suggestion: add new dithering method for those who are willing to experiment and test it. The new method will limit the distance of dithering points from the starting point by a certain percentage of camera FOV (as my test shows, there is no such limit in place) while making sure that each dither point is still unique. I’m not claiming expertise in dithering algorithms, so I'm reaching out for input on possible enhancements for this new method. The goal is to maximize the efficiency of the limited imaging time we have. Optimal dithering patterns might vary based on sensor characteristics, such as pixel size and noise level, as well as other factors like the need for drizzling.

Thanks,

Leo