From the person who spent some money and enormous time on lots, lots, lots and lots of torubleshooting of composite/chroma using datasheets, reference designs, multimeter and oscilloscope.
And still having no definite answer what the issue actually is, but suspect the issues are in timing (which is controlled by VDP and we can not change unless connect advanced decoding/encoding device). In respect to my problems with AD725, I decided to abandon this chip (which works @ 4x speed to the MSX) and use CXA1645. It encodes RGB picture more or less correct my LCD TV displays. Looks bad in 80-character mode though.
During the course of actions, I have learned the following:
- CRTs and modern TVs may have different voltages (Vpp) for displaying colors. MSX displays accept 1 Vpp (peak-to-peak), while modern TVs are having white at 0.741 Vpp. The diffence can be very well seen doing COLOR 15,14,14 - white chars on light gray background: my TV displays almost white background,a dwhite text is hardly seen. The same story may go for composite output: its Vpp voltage may simply be out of spec for modern TVs.
- Timing. Modern TVs, encoders and decoders may be very critical to the duration of the SYNC signal, and position of the color burst in the output. They may be "too clever" having false positives in the signal output by the MSX. They may have issues syncing with the signal for these reasons, having dot crawl, jumping characters, or color artifacts. And they worke on higher clock speeds to decode signal - while our common sense tells us that higher speed/higher resolution is good, it has downside as false positives or negatives.
- Everyone says that, but I did not believe: composite is generally a bad choice for text and high resolution graphics. Text has sharp freq/amp level changes, leading to noise (thus artifacts). When I connect my Yamaha PC to the 40-inch TV I see exactly the first picture you posted. At least it is static, however having artifacts. My machine is having "color/BW" switch at the back, if I switch to BW mode (luma only) I get excellent grayscale picture.
I may have narrowed down the color issue: the video capture software I was using may have not been initializing the capture devices correctly, and it wasn't exposing all of the available driver settings.
I put an oscilloscope on the video output from the CX5M, and discovered that the signal was 1.24Vpp (volts peak-to-peak). I believe the standard states that it should be 1.0Vpp. It's possible that the color issue was a lack of gain control (or non-enabled gain control) on the capture devices. I have ordered an attenuator, thanks to finding one on eBay at a fair price, in an attempt to determine if that was a cause.
Another forum (The Digital FAQ) recommends trying VirtualDub to capture video, as it exposes settings that other capture software may not. This actually worked with one of my capture devices: I was able to capture nearly-correct colors without having to change any settings. (Orange was slightly too red.) I am still trying to determine what VirtualDub is doing differently.
And I also ordered an external comb filter to try with is capture device... eBay to the rescue again.
The majority of the 80s computers/consoles, as a measure to save cost/energy, generated the CVBS signal using a trick called composite artifact colors. But, for that to work, it requires good filtering on the receptor side.
The conclusion is that your capture cards/TVs that are showing vertical bars on the CVBS input just have crappy filters on their input, so the vertical bars end up showing.
Ah yes, I now remember reading about composite artifact colors. Looks like I have many capture devices with poor filtering. (Which is kind of sad, because one of them was supposedly a top-of-the-line consumer card in its day, and probably cost $200 or more when new.)
Oh, right, but NTSC TVs have a tint control, don't they? ... So do you definitely have the same tint value on both devices?
My CRT TV is at its default setting. I left the capture devices to the same. Since NTSC doesn't specify how tint setting values are conveyed, I can only say that they all default with the setting in the middle.
Definitely have the same tint value? No.
Of course, the trade-off of fewer channels fitting in the available broadcast space was never really a problem in my country, being an island with a decent amount of government control of broadcasting. So that's lucky!
Only one of our states is an island... I guess we needed the broadcast density.
I also experienced these issues designing the GFX9918 card of the Artemisa MSX computer.
In fact, could you make an interesting experiment with your TMS9918? Try to execute the following Basic code:
10 SCREEN 2 20 VDP(7)=0 30 A$=INKEY$ 40 IF A$="" THEN 30
This just sets the VDP in screen 2 and selects the backdrop color (screen border color) to 0 (transparent). According to the 9918 specs, this should fallback to black color. But in some TV sets, this causes the synchronization to be lost. I think this might be because the transparent color is slightly below black color, and some CVBS decoders detect this as a synchronization pulse.
Could you try that with your 9918-based computers?