Hello!

I've spent the last 12 hours trying to figure out how to derive the RGB values of the MSX1 color palette from the TMS9918 datasheet. I couldn't find any clear indication on how to calculate the RGB values from the Y, R-Y and B-Y values given in the datasheet. The values are given as percents of the black/white voltage swing, and it doesn't help me to figure out how do these values translate into (possibly) Y'PbPr colorspace, so I can apply the proper transformation matrix.

I have already found one derivation here:
http://bifi.msxnet.org/msxnet/tech/tms9918a.txt

Which by the way are the same values used in BlueMSX. However, this calculation derives the RGB values without taking into consideration the whole transformation matrix, as given here:
http://en.wikipedia.org/wiki/YCbCr
and here:
http://www.poynton.com/notes/colour_and_gamma/ColorFAQ.html#RTFToC28

Anyway, I couldn't figure out how to translate de relative-voltage values. Does the 0.47 black level means that I have to subtract 0.47 from every value in order to get them to the [-0.5, 0.5] range defined for this color space? What happens with a value of 1.0, like the Dark Blue color? Should I instead subtract 0.5? Would it mean that black is not completely black?

I believe the key to understanding this is to understand how outputs signals Y, B-Y and R-Y are processed in a RGB encoder or in a monitor. However, I lack the necessary hardware knowledge. Can anyone help?

Thanks,
Leonardo

Here's how openMSX does it.

I can see that openMSX uses the exact same palette as blueMSX.
I've done my own calculations, and arrived at these values:
0, 0, 0
0, 241, 20
68, 249, 86
85, 79, 255
128, 111, 255
250, 80, 51
12, 255, 255
255, 81, 52
255, 115, 86
226, 210, 4
242, 217, 71
4, 212, 19
231, 80, 229
208, 208, 208
255, 255, 255

What I did was to apply gamma correction, so that the 0.47 from the datasheet would fall at 0.5 without changing the 0-1 extremes, and used the proper YCbCr conversion, considering that the values were normalized. Also, I clamped resulting values below zero and above 255.

What do you think about this palette? I believe my yellow tones are much better ;-)

--
Leonardo

...waiting for _Wolf to comment

Oo?

I'd go for the Mexican color scheme any day..

¡Yo no lo compreendo, señor!

never mind, inside joke

I'll give your palette a try Leornado. blueMSX uses indeed the derivation you mention. I'll connect my real MSX and emulator to the same monitor and see which palette looks closest to the real MSX.

depends on both your monitors and NTSC vs PAL I suppose.... It's really tough to make such a decision. OTOH: it's nice to see the values from a different approach.

Quote:

never mind, inside joke

Yeah, that 1 never grows tired. hahah

I've just read the article about the YUV color space, and redone my calculations. I was using the YCbCr transform, now I tried the YUV transform:

0, 0, 0
38, 229, 0
100, 240, 49
93, 57, 255
132, 94, 255
214, 104, 20
64, 228, 255
253, 111, 15
255, 145, 49
216, 230, 0
232, 233, 14
40, 202, 0
205, 86, 255
208, 208, 208
255, 255, 255

It felt weird, though. The red tones seem a little bit too orange...

Serves me right for doing things in a hurry... I made a mistake, now it's correct with regard to YUV color space. These are really cool!

0, 0, 0
0, 241, 20
68, 249, 86
85, 79, 255
128, 111, 255
250, 80, 51
12, 255, 255
255, 81, 52
255, 115, 86
226, 210, 4
242, 217, 71
4, 212, 19
231, 80, 229
208, 208, 208
255, 255, 255

Leonardo

OMG... the last values are the same with regard to YCbCr colorspace...
I triple checked it: i used different matrices to make the calculations. Maybe it means something?

Like manuel noted, it depends on the screen (monitor/tv) decoder, and also user-preferred hue, saturation, brightness, contrast, gamma. There is no single "best" MSX1 palette. The matrices you're calculating with are probably based on standard NTSC.

Somewhat related, an old discussion about custom MSX1 palettes: http://www.bluemsx.com/blueforum/viewtopic.php?t=1281
If (to make mrbolha happy ) emulator authors are planning to implement user-customised MSX1 palettes, we should use a standard for inter-compatibility. I suggest a binary 45 byte .pal file, with as contents 24-bit r,g,b data of the 15 colours.