TMS9918 Video Output Behavior?

Here are some examples of #2 as mentioned in my first post above, using my National CF-2000 (my CX5Ms were downstairs, and the results would look the same):

Using a Diamond Multimedia VC500 via its Composite input:

Using a Diamond Multimedia VC500 via its S-Video input*:

Using a Avermedia Easy DVD 7 (C039) via its Composite input:

Using a Avermedia Easy DVD 7 (C039) via its S-Video input*:

(OK, this one doesn't look too bad... the blue color is a bit off, but the vertical pattern is even.)

* S-Video inputs were using a StarTech Composite to S-Video adapter.

I purchased a RetroTINK-2X, and that seems to have eliminated the vertical lines I am seeing in the video capture. That would seem to indicate that the vertical lines are an artifact of 240P output when the input does not support that resolution.

However, I am still seeing an unexpected color shift -- the MSX color Dark Red looks orange and MSX color Dark Blue looks muted. (The two yellow colors also take on a green hue.) I'm also suspecting the MSX video cable I purchased is not a 75 ohm cable.

Since I am trying to document the CX5M specifically (the longer-lived of the two MSX computers released in the US), I stand behind my decision to try to use the actual hardware. However, this exercise is giving me a headache in the limitations in the NTSC video standard verses improvements made to other standards that were devised afterwards...

The vertical lines do indeed betray a deliberate deviation from the NTSC standard: the TMS produces in-phase video.

NTSC adds colour to the existing black and white signal via a colour subcarrier — a high-frequency and low-intensity signal that is added atop the black and white. Both of those features are meant to make it hard to perceive on an older or cheaper black and white set that doesn't filter it out, but so is a third feature: relative phase.

The frequency of the colour subcarrier is selected so that 227.5 cycles of it fit on every line. So the colour content on each line is 0.5 cycles offset from the lines on either side. So in an area of consistent colour, the peaks and troughs in the colour signal are in between each other, line by line. Furthermore, because there are an odd number of lines in a field, they alternate from one field to the next.

All of that helps to make them much less obvious to the eye.

As per the exact verticals, the MSX doesn't do that. It deliberately outputs lines that are fractionally too long: long enough that they contain exactly 228 colour cycles. That's a great simplification for digital video generation as it means no state needs to survive from one line to the next, and parts of the signal are easier to generate — especially the colour burst (an invisible part of the signal in the horizontal border that provides colour phase information) is now identical on every line and therefore not a function of anything. Just a fixed bit of signal to insert.

Some other machines do the same thing: all the 8-bit Ataris and early C64s spring to mind.

This can cause visible artefacts because there's no such thing as a perfect chroma/luma filter — a perfect way to separate colour and brightness information that have been combined in a composite signal. So some decoders will hedge their bets by allowing small amounts of the colour signal to survive in the brightness part, rather than more aggressively filtering the brightness in a way that would reduce sharpness. The trace amounts would be hard to perceive anyway, were the signal on-spec and including the difference in phase per line.

The same machine connected to a CRT TV.

Note: PAL does not really correct any errors. That was a misleading marketing claim by Telefunken that stuck without questioning.

What PAL really does is:

1) Reduce the vertical color resolution by half
2) Replace the hue errors by saturation errors, which is less noticeable.

But the trick had it restrictions. More severe errors would just translate into Hanover bars.

Also, in direct CVBS cable transmission from the computer/console/DVD to the TV, phase errors are negligible. In this case, all that PAL does is to worsen the image quality by halving the vertical color resolution.

I'm not criticising, just explaining. This was the era of nifty tricks with very restricted hw resources. They did the best they could with what they had at hand.

But what doesn't make any sense is people bashing NTSC nowadays for direct CVBS cable connection, when PAL performs worse in this application.