nesdev.com

http://www.neo-geo.com/forums/showthrea ... NES-Thread

Doesn't the "stock NES composite" screenshots look way too good to be composite video, especially for an NES? What is in that guys tv, that filters out rainbow and dot crawl artifacts so perfectly, and at the same time is clear enough to see every pixel?

That doesn't look like the stock composite output at all.

BTW, speaking of scanlines, are they as pronounced on anyone's TV as in these pictures? I never got this obsession with scanlines that emulators and filters have, since none of the TVs I have played video games on since I was a kid had noticeable scanlines.

No, absolutely never in my case. I think the only time I ever saw scanlines on a CRT TV was when I lowered the settings so much that white was near black (the color was so dark that you'd see scanlines, and even then it was on a screen nearly full of that color). So yeah.

Same goes with curvature. Like WTF people, I never saw a CRT that was so curve, the deformation was so mild that it may as well not be there for practical purposes (you'd really notice only if you got really close, like right next to it). I think the only reason CRT emulators tend to curve images so much is because otherwise one thinks it isn't there. But then again, I used early '90s CRTs, maybe earlier ones were more curve.

In the "stock composite" shot of the hill behind Mario's back, the line between the dark green and black has the telltale 3-line roughness pattern. I wonder if the camera's built-in processing is filtering out some of the rainbow artifacts from thin vertical lines, or if a 33 ms exposure is averaging the even and odd phases of the color sequence.

Perhaps some cheap TVs were designed for a softer focus of the electron beam, which would cause less of a "scanlines" effect than seen here. But TVs and monitors designed to show the full detail of 480i (interlaced) video will tend to focus their electron beams sharply enough to make even lines of even fields and odd lines of odd fields distinguishable. This is especially likely in monitors with an RGB, component, or S-Video input, as those signals tend to carry more picture detail than composite. This focus causes gaps between lines in 240p video. But I've never really noticed "curvature" as some plug-ins call it.

I wonder if he's using a CRT HDTV...

The only CRT HDTV that I've used upsampled 240p and 480i to 480p, probably because Samsung didn't want to make a tube that can sync at both the 15.7 kHz of SDTV and the 31-34 kHz of EDTV and HDTV. This TV appears to be actually displaying 240p.

Maybe the TV just overlaps consecutive frames at their closest 180 degree phase match. That way, 3/4 of a pixel would be the correct color.

Every CRT EDTV/HDTV I've used works this way. I'd be very happy to find a truly multi-sync CRT TV, but it's not something you just come across.

I also see the 3-step jagged edge pattern, it is easily seen in the grey triangular bush bit behind mario. Super Mario Bros 3 is very good at making composite look less awful than it usually does, I think largely because of the black outlines which prevent different hues from meeting much of the time. I've used some mid-2000s CRTs that had fairly good comb filters that eliminated a lot of the horizontal color artifacting. In the process, though, they would often introduce dots on straight horizontal lines, an interesting irony considering that usually one thing composite can transmit without much distortion.

Sony TVs and multisync monitors both have very narrow scanlines, so I'm guessing one of those categories. Also, when taking a picture of a CRT, I often had to either set the aperture manually on my camera, or reduce the brightness on the TV/monitor. Otherwise it's an overloaded mess. Thus even a regular TV with thick scanlines could look like those photos at a reduced brightness.

_________________
http://www.chrismcovell.com

It's probably a Sony PVM broadcast monitor (not a typical consumer set). http://retrorgb.com/rgbmonitors.html

I am talking about the composite quality.

The PVM is still relevant. They have very good composite quality compared to many contemporary televisions. Some PVM monitors are composite only, even.

Slightly off-topic, but I thought of an easy way Nintendo could've reduced dot-crawl on the NES. Add a negative chroma signal, delayed by 90 degrees, to the composite signal.

Wut?

Composite = Luma + Chroma
Chroma = U·sin(t) + V·cos(t)
"Delayed Chroma by 90 degrees" → U·cos(t) - V·sin(t)
Composte + "Delayed Chroma by 90 degrees" = Luma + U(sin(t)+cos(t)) + V(cos(t)-sin(t)) = Luma + U·sin(t+π÷4)÷√2 + V·cos(t+π÷4)÷√2

That would just increase the saturation of the chroma signal.

To minimize vertical and diagonal artifacts on NTSC, you want to put successive lines 180 degrees out of phase. The master clock of the NES and Super NES is six times the NTSC color burst frequency, so you want to do an odd multiple of 3. The PPU ordinarily generates a scanline 1364 pixels long. Lengthening it to 1365 (455*3) would cause the positive and negative phases of the I and Q (or U and V) subcarriers on each scanline lining up below each other, making vertical lines much less ragged. Because the 3 is not a multiple of a dot (4 pixels), diagonals would be ragged, but they would be evenly ragged, unlike on the NES where a \ diagonal is even and a / diagonal less so. (Check the hills in SMB1.) Checkerboard dithering (e.g. Solstice, Dr. Mario) would produce a different, possibly less objectionable pattern as well.

With single fields taken care of, we now need to alternate the dot pattern from field to field. Changing the phase by three master clocks by dropping the 1365th master clock from three lines each vblank would do it.

Moreover, this change to composite signal generation could be implemented in a NOAC without sacrificing compatibility by having the PPU stall the CPU's divide-by-12 circuit for one master clock on each scanline.