And now... averaging V' over pairs of lines in PAL.
It's interesting for sure: you can certainly configure it so that alternating red/blue gives you a nice "purple" (e.g. in the Superkid loading screen). The available filters may need work though, or maybe the averaging algorithm shouldn't be so straightforward: Donut Dilemma shows an odd colour flash atop your bonce while you jump...
robcfg wrote: ↑Sun Mar 26, 2023 10:43 pm
How are you averaging the lines? Does it work consistently on graphic modes and semigraphics? They seem to be offset one line.
If should be averaging only the V' chroma component of each line (with Y' and U' remaining distinct per line). Not sure what you mean by an offset, but it does look like some lines end up with parts "going missing".
Edit: BTW, you can get the "nice purple" by setting encode filters on "I and Q" (actually U and V for PAL) - 1300 should be a nice number, whichever windowing function you choose. This low-pass filters each chroma channel before it's added to the signal.
More edit: so I've made that the default for PAL. Also, averaging three parts current line with one part its PAL pair seems to help out with those disappearing elements a bit. At least on the Superkid loading screen
Well it's not there yet! But a good link - reading through that thread reminded me about the way the Dragon is "not quite centred" with its phase alternation, which may lead towards a better implementation. Of course it'll need more code and preferably for me to break out the oscilloscope again
Just pushed out another build that lets you configure custom sinc windowing function and corner frequencies - so you can continuously vary the cutoff to see what works. The other precalculated options are still there too.
You can also adjust how much V' is taken from the "alternate line" in PAL. However I still can't make Pere's pictures look good