Oops, just kidding–the S-Video splitter I was using was wired weird and had C coming through somehow.
Used a different splitter where it was definitely only Y coming through and now it looks great.
I’ll do some more tests shortly and post some clips.
I must admit I haven’t been following the technical details of this thread closely, but if overall luma is affecting your sync situation then the underlying problem could have something to do with your composite signal’s effective average DC offset (a white screen is basically a positive DC offset, minus sync etc). Something something grounding something something black level restoration something something DC decoupling. Wow, that’s about the vaguest I think I’ve ever been in my whole life… sorry 
You would very likely benefit from some detailed scope work to determine exactly what’s going on and get some more info.
No that’s totally fair considering how vague my descriptions are compared to providing actual measurements with a scope. All good reminders on what I should be checking while breadboarding this up.
Right after I posted that, I realized I was accidentally using a S-Video adapter that merged Y/C in a RCA jack which happens to look identical to another adapter I have that only pulls Y out. Using just Y completely got rid of the diagonal line issue I was describing above.
In any case, the YPbPr output now looks pretty clean apart from a small but still perceptible amount of noise on black which could be the breadboard or cheap cables or maybe just my CRT!
Comparing the dot crawl and overall quality I was getting on composite with YPbPr output is a night and day difference–so much sharper. Will post some captures soon. Thanks for the idea on using that signal @Fox.
We had discussed technical details a bit further in DMs as well, but I guess the brief version of what I suspect has happened is that the sync injection stage creates an offset when sync is high. To get around this we need to use an open collector transistor connected to a negative voltage with an appropriate resistor which will drop Y below 0 during sync, but be high Z when sync is high. That is assuming we continue with the BA7230LS option.
If we do though, I plan on drawing a circuit to inject sync properly but I first have to decide what negative voltage to use and calculate the passive components from there. We could just use -12v since it is available, but we were already discussing using a dedicated +5v vreg for further noise isolation so why not use an additional -5v or -3.3v regulator too?
Buuuuut, we already have an acceptable Y with sync from the AD724 so we may be able to simplify the circuit so much further by doing the B-Y and R-Y operations.
Glad it got worked out. I know that we can continue to simplify and improve the circuit with a working prototype. I’ve been learning a lot from all of this reading.
I just read something interesting on the YPbPr wikipedia page:
I think that explains the scrolling checkerboard you described when you had chroma mixed in with luma, although this is irrelevant to the final goal.
I tried this by putting just the Y plug into the composite out and it gave the exact same diagonal line pattern as yesterday–good to know that’s chroma interference if I see that again!
Comparing Fabio’s board with Dek’s, I find it hard to believe a fab house even printed Fabio’s. There are several vias within vias. The BA7’s datasheet also demands users tie unused inputs pins to ground via a 1uF capacitor which Dek’s board luckily does. Pins 3, 4, 5 and 20 were left floating, and while the datasheet doesn’t elaborate on possible negative effects, it’s usually drift, self oscillation and noise.
Bodging some caps on these pins might help with the noise on black if you still feel like bothering with the first board.
I did some captures to test the difference between composite, S-Video, and YPbPr as well as putting them through a Retrotink 2x for an HDMI out, all going into an Intensity Shuttle.
Just for the YPbPr outputs:
When I run the YPbPr output directly into the Shuttle, I get this weird strobing effect that scrolls downwards.
Color output:
Black output. You can still see the strobing faintly if you look closely. It’s more obvious on my computer monitor while capturing so YouTube might be squashing it:
If I run the YPbPr output into the Retrotink 2x and then into the Shuttle’s HDMI input, the strobing almost completely disappears and there’s very little noticeable noise. Probably can still be improved.
Color output:
Black output:
I’m not certain how this is changing anything other than just transcoding the 480i signal into an HDMI compatible signal. Maybe I’m getting reflections somewhere and the Retrotink filters them out?
I put some 1uF caps on those pins and didn’t notice any immediate difference in quality on my CRT but forgot to capture test footage.
Looking forward to getting this built with the new PCBs when they come in next week.
I can hardly see it because of youtube’s compression. But I can see some effect that isn’t present in the second color video. I can’t imagine what it could be due to. To me it almost looks like some fluctuation on the brightness signal and even then it is most notable on cyan and blue. Covering the left half of my screen, its virtually gone from the green/black stripes.
Again I assume this is because of youtube’s compression, but the video is perfectly black on my laptop screen.
Ok. Improvements are probably too subtle to see, but still recommended either way. Always good to cover all bases.