TV Emulation doesn't have accurate colors

[Michaelangel007]

Spun off from #357

Source

#254 Gives this program

10 HGR
20 FOR Y=0 TO 63:FOR X=0 TO 7
30 HCOLOR=INT(Y/8)
31 HPLOT X*32,Y*2 TO X*32+30,Y*2
32 HCOLOR=X
33 HPLOT X*32,Y*2+1 TO X*32+30,Y*2+1
40 NEXT:NEXT

1.25

1.25 Monitor mode:

1.25 TV mode:

1.26

1.26 Monitor mode:

1.26 TV mode

Real TV

On a real TV green + orange look yellow, matching 1.25 more:

Notes:

• This is NOT a CRT -- it is a Sony Bravia LCD
  • It is hooked directly to the TV via the composite input
  • The yellow on the TV is brighter then 1.25
  • I do not have first-hand access to it.
• I do have four Composite-to-VGA adapters on order from Amazon in the hope that I can replicate this:
  • SLLEA RCA Composite AV S-Video to VGA Converter Box
  • ESYNIC CCTV Camera BNC S-Video VGA to VGA Converter Box
  • WSTD PC Laptop Composite Video TV RCA Composite S-Video AV In To PC VGA
  • Portta AV/CVBS RCA Composite to HDMI Mini Converter
• We still need a picture from a real CRT TV (using a RF modulator??)
• Regardless of the output (CRT vs LCD) the TV mode in 1.26 regressed from 1.25

[Michaelangel007]

Source 2a

Let's explicitly render these 14 "fake" HGR colors sorted by hue:

1 GOTO 10
2 X=I*14:Y=J*16
3 FOR V=0 TO 7:T=Y+V*2
4 HCOLOR=A:HPLOT X,T TO X+13,T
5 IF S=1 THEN 9
6 HCOLOR=B:HPLOT X,T+1 TO X+13,T+1
7 HCOLOR=A:HPLOT X,T+40 TO X+11,T+40
8 HCOLOR=B:HPLOT X,T+72 TO X+11,T+72
9 NEXT:I=I+1:RETURN
10 HGR:S=1
20 A=5:B=0:I=2:J=0:GOSUB 2
30 A=1:I=5:GOSUB 2
40 A=6:I=8:GOSUB 2
50 A=2:I=11:GOSUB 2
60 A=3:B=3:I=13:GOSUB 2
70 I=0:J=1:S=0
80 A=2:B=5:GOSUB 2
90 A=5:B=7:GOSUB 2
100 A=5:B=5:GOSUB 2:REM ORANGE
110 A=5:B=1:GOSUB 2:REM yellow
120 A=1:B=3:GOSUB 2
130 A=1:B=1:GOSUB 2:REM GREEN
140 A=6:B=1:GOSUB 2
150 A=6:B=7:GOSUB 2
160 A=6:B=6:GOSUB 2:REM BLUE
170 A=6:B=2:GOSUB 2
180 A=2:B=7:GOSUB 2
190 A=2:B=2:GOSUB 2:REM MAGENTA
200 A=1:B=2:GOSUB 2:REM gray
210 A=3:B=3:GOSUB 2:REM WHITE
220 HCOLOR=3
230 FOR I=0 TO 13
240 HPLOT I*14+6,40 TO I*14+10,40
250 HPLOT I*14+6,42 TO I*14+10,42
260 HPLOT I*14+6,76 TO I*14+6,80
270 HPLOT I*14+4,78 TO I*14+8,78
280 NEXT

1.25 TV mode

1.26 TV mode

The vertical mixing of two adjacent scanlines looks worse in 1.26 then 1.25.

Now instead of having a full 256*256 lookup table, we should be able to get with a much smaller table.Why? If we treat the HGR screen as only producing the valid 16 GR / DHGR colors we only need to have a [16][16] mixing table.

We still need to work out how the bits need to be mixed. i.e. Do we take the top 4 bits of RGB ?

Solutions

We can either:

• pre-process in YIQ space
• process-in-place in 4-bit / 12-bit space
• post-process this in RGB space

Which one is more accurate vs faster is TBD.

[Michaelangel007]

Updated all images to be correct.

TL:DR;

• 1.25 TV mode colors are TOO solid
• 1.26 TV mode colors are TOO pixelated

For 1.27 we want to aim more in the middle.

Unanswered questions:

• Can we just scan-double the output lines once we have the correct blended line?
• Can we just do a linear (line1+line2)/2 and replicate to both lines?
• Can we use a 3-way tap three times??

Let's suppose we want to generate TV mode for lines N, and N+1

N-1| line 0 | A 
N+0| line 1 | B  \ TV mode mixing
N+1| line 2 | C  / TV mode mixing
N+2| line 3 | D

Does it blend? There are multiple ways we can blend:

Blend 1

line 1 = (A + B) / 2
line 2 = (C + D) / 2

Blend 2

line 1 = (B+C)/2
line 2 = (B+C)/2

Blend 3

line1 = A/4 + B/2 + C/4
line2 = B/4 + C/2 + D/4

[sicklittlemonkey]

1.25 TV mode colors are TOO solid

Agreed.

1.26 TV mode colors are TOO pixelated

The AppleWin 1.26 screenshots you have above are actually in Color Monitor mode. Shouldn't you be comparing Color TV mode? (This is accurate as to how I remember yellow looked in RGB and PAL composite.) Cheers, Nick.

[Michaelangel007]

If we use a naive bicubic down-sample Program 2 TV mode:

.. 50%:

... 25%:

And then bicubic re-up-sample back up to 560px, we get these solid colors:

We get similar results if we bilinear downsample / upsample monitor mode:

Conclusion

A naive re-sampling matches the hues of the actual TV pretty good.

[Michaelangel007]

Let's convert program 2 from HGR to GR:

Source 2b

1 GOTO 10
2 X=I*2:Y=J*4
3 FOR V=0 TO 3:T=Y+V
4 COLOR=A:PLOT X,T:PLOT X+1,T
5 IF S=1 THEN 9
6 REM ---
7 COLOR=B:PLOT X,T+10:PLOT X+1,T+10
8 COLOR=C:PLOT X,T+18:PLOT X+1,T+18
9 NEXT:I=I+1:RETURN
10 GR:S=1:J=0
20 A= 9:I= 2:GOSUB 2:REM ORANGE
30 A=12:I= 5:GOSUB 2:REM GREEN
40 A= 6:I= 8:GOSUB 2:REM BLUE
50 A= 3:I=11:GOSUB 2:REM MAGENTA
60 A=15:I=13:GOSUB 2:REM WHITE
70 I=0:J=1:S=0
80  A=11:B= 3:C= 9:GOSUB 2
90  A= 0:B= 9:C=15:GOSUB 2
100 A= 9:B= 9:C= 9:GOSUB 2:REM ORANGE
110 A=13:B= 9:C=12:GOSUB 2:REM yellow
120 A=14:B=12:C=15:GOSUB 2
130 A=12:B=12:C=12:GOSUB 2:REM GREEN
140 A= 0:B= 6:C=12:GOSUB 2
150 A= 0:B= 6:C=15:GOSUB 2
160 A= 6:B= 6:C= 6:GOSUB 2:REM BLUE
170 A= 7:B= 6:C= 3:GOSUB 2
180 A= 0:B= 3:C=15:GOSUB 2
190 A= 3:B= 3:C= 3:GOSUB 2:REM MAGENTA=3
200 A= 5:B=12:C= 3:GOSUB 2:REM gray
210 A=15:B=15:C=15:GOSUB 2:REM WHITE
300 FOR I=0 TO 15:COLOR=I:PLOT I*2,28:NEXT

Output

I've set the GR colors that don't match the fake HGR colors as black:

Monitor Mode

TV Mode

[Michaelangel007]

@sicklittlemonkey Thanks Nick - - images should be correct now. Can you verify please? Thanks

[Michaelangel007]

Disk image (ProDOS /HGR.TV) with source programs:

hgr.tv.dsk.zip

[6502-workshop]

Thanks for posting this Michael! I am using colors affected by this issue extensively in Nox Archaist.

[Michaelangel007]

No prob Mark.

Updated comment 1 with extra information -- yellow, and and Composite-to-VGA adapter/convertors