Stella Gets Some New Clothes

[Thomas Jentzsch]

Here are the pictures

• Screenshot of my LCD with Stella (for calibration)
• Optimal
• Various degrees difference to optimal
• B/W (to show the darker grays, the brighter top is from the TV scanning)
• 2x in large, to show my problem (unfortunately my camera doesn't display the bright colors well here, so you don' t see the red tint very well)

Hues Cx and Dx are indeed close and least affected (especially visible in the Minus_20 and 30 shots).

Edited October 2, 2013 by Thomas Jentzsch

[SeaGtGruff]

Thank you very much, Thomas! But I've just got to ask-- what time zone are you in? Is it November 8th there, or the 11th of August?

 

I'm going to see if I can adapt my NTSC palette calculations in line with how I think the PAL colors are being generated, to see if I can duplicate these. Personally, I think I'd classify the "Minus_10" picture as pretty close to optimal-- maybe not the closest to the Stella screenshot, but preferable because it looks like it includes a pretty good yellow and otherwise has a nice range of colors.

[Trebor]

Awesome captures, Thomas. I would have to respectfully disagree with SeaGTGruff and say the Optimal appears Optimal as opposed to a -10 degrees...

 

-10 degrees follows more of how colors would look under an LCD/LED/Plasma; the optimal looks like the system on a CRT; at least from my experience.

Looks like what would be the equivalent for NTSC 9x almost doubles as PAL Bx and Dx, with perhaps a tad more red under Bx (?)

 

A quick once over looks like (If we were arranging NTSC colors for it):

0x
0x
1x
Dx
3x
Cx
5x
Bx
6x
Ax
7x
9x
8x
9x
0x
0x

...Around those lines. Looking forward to what SeaGTGruff come up with...Awesome work by all!

[Thomas Jentzsch]

Looks like what would be the equivalent for NTSC 9x almost doubles as PAL Bx and Dx, with perhaps a tad more red under Bx (?)

No, Dx is (very!) slightly more red. Best noticable towards the brighter colors.

[Trebor]

Thanks for the clarification...I would need to study the captures longer on better equipment as I'm at work now and not under the most ideal display and conditions for viewing the captures.

 

Thanks again though for the captures. A really good/'ideal' PAL palette should be obtainable from them.

[Trebor]

Thomas,
I think I see what your PAL system is doing now (Better under an IPS monitor at home). It's less not more than the NTSC system regarding phase shift. I'm presuming, like NTSC there will be a 0.5 degree phase shift between a cooler and warmer system, but instead of going forward, I think we need to go backwards with PAL. I understand that now - thank you again for those excellent pictures. I'm going to work on Stella242, Stella237, Stella232, Stella227; for PAL values only though.

There's likely a pairing that will go something like this: NTSC247 & PAL232 (Cold); NTSC252 & PAL237 (Warm). Give or take half a degree for PAL. You'll be able to confirm more exact the PAL value.


SeaGTGruff,
If you're building off of NTSC, I believe you want to place the NTSC colors in the following order utilizing the below:

0$
0$
2$
D$
3$
C$
4$
B$
5$
A$
6$
9$
7$
8$
0$
0$

Utilizing the above order off of NTSC colors with a Phase Shift in the vicinity of 22.7/23.2 thru 23.7/24.2 degrees may do "it" in that case.

[SeaGtGruff]

Thomas:

 

I modified my NTSC palette calculations to compute the PAL colors using my theoretical model. I can see that my model is definitely off somewhere, although it's clear from your screenshots of different pot settings that the least phase shift occurs with hues 12 and 13, and the most phase shift occurs with hues 2 and 3. It amost looks like hue 12 is the "zero shift" color, rather than hue 13 as I'd thought, but I'm not sure how that works. I guess I first need to understand the NTSC phase-shift mechanism better. The way the color values are decoded-- i.e., the order they're in with respect to the delay line that runs through the circuits-- I'm led e to believe it's an additive process. I assumed there must be some comparisons going on, although I haven't tried to puzzle out how those circuits work yet. That is, if the delay line is going into hue 15 first and into hue 2 last, I don't see how the shifting can be cumulative unless any given hue gets the shift for itself plus the shift for all hues less than it, but not the shifts for any hues greater than it:

 

For hue 15, total shift = hue 15's shift + hue 14's shift + hue 13's shift + ... + hue 2's shift

For hue 14, total shift = hue 14's shift + hue 13's shift + hue 12's shift + ... + hue 2's shift

etc.

 

To me, that implies some sort of comparison, because if, say, the hue 12 decode line is the only one that's active, then how can the shifts for hues 11, 10, 9, etc. get added to the shift for hue 12 when those decode lines aren't active? But maybe it's some kind of cascading effect. Now that I think about it, maybe the delay signal is allowed through whenever a hue's decode line is inactive, but an active decode line stops (blocks) the delay signal, and once it's stopped it remains "off" for the rest of the hues' circuits? I *really* need to sit down and figure out how that part of the schematic works.

 

I had assumed that the hues need be arrayed in a numerical order (either increasing or decreasing) with respect to the delay line, which is why I figured hue 13 must be the last in line. But if there's no sort of comparison going on, I guess the order doesn't matter except for determining how much shift gets applied for any given hue. I wish someone (Curt most likely) can eventually manage to locate the PAL version of the TIA schematics so we won't have to guess.

 

Anyway, here are some PAL palettes as computed according to my faulty theoretical model.

 

With a phase shift of 360 / 23, giving the most "ideal" or evenly-spaced colors, looking similar (but not identical) to your "Minus_10" screenshot:

 

 

With a phase shift of 360 / 26, which looks similar to Trebor's "Optimal" palette:

 

 

With a phase shift of 360 / 37, which looks similar to your "Plus_5" screenshot:

 

 

And finally, with a phase shift of minus 27 degrees, which looks somewhat similar to your "Minus_30" screenshot:

 

 

You can see that despite the similarities, there are also noticeable differences, which indicates that my model is incorrect.

 

Trebor:

 

I don't think that using the NTSC palette with the hues rearranged, and with different phase shift values, will be sufficient, as that method doesn't account for the fact the hues 12 and 13 change the least, and hues 2 and 3 change the most, as the color pot is adjusted. But there's no denying that your "257" palette matches the "Optimal" screenshot that Thomas posted, whereas my attempt (second screenshot above) clearly differs in the "center" of the palette (purples and blues).

 

[Thomas Jentzsch]

Before you optimize the palettes, I hope you noticed that my "Optimal" screenshot has a slight cyan tint.

 

Most noticeable is the difference at Ax which looks much more red on screen than on the photo. Cx is also still a bit more purple than on the photo.

[Trebor]

Thomas,

 

Please try these: Stella227-242PALONLY.zip

 

My thought is perhaps Stella227PAL is your system 'cold' (Under ~30 minutes consistent runtime), and Stella232PAL is your system 'warm' (Over ~30 minutes consistent runtime). It could be off by half a degree, and obviously it is no problem to go lower. I've included 237 and 242 to complete the chain (22.7-27.7 degree phase shifts in 0.5 degree increments).

 

However, I've only created the PAL palettes as indicated in the file names. The original Stella NTSC palette is the placeholder until I work on the NTSC palettes for the aforementioned phase values. Then again, it probably makes better sense to couple them the way I alluded to in my prior post.

 

Here are some RGB captures...

'Default' 50 Sat --> 75 Sat --> 75 Sat + 75 Gam:

 

PHASE SHIFT 22.7:

PHASE SHIFT 23.2:

PHASE SHIFT 23.7:

PHASE SHIFT 24.2:

[Trebor]

To show what that half a degree (less) looks like, and just in case it better matches, you can grab the PAL 22.2 degree phase shift here: Stella222PALONLY.zip

 

Captures matching the previous post order:

 

I'm still thinking towards the 22.7 degree phase shift as 'cold' and a 23.2 degree phase shift as 'warm' for PAL, but what the PAL hardware shows is of course what really counts.

 

Thanks again, Thomas!

[Thomas Jentzsch]

Something still doesn't match.

 

E.g.

• Cx looks closest with a high phase step like 26.7°,
• 9x comes closest with 23.2°.
• 7x: 26.2°
• 6x, 8x: 22.2° (maybe even slightly less)
• Dx: 27,2°

Pretty weird.

[Trebor]

 

Something still doesn't match.

 

E.g.

• Cx looks closest with a high phase step like 26.7°,
• 9x comes closest with 23.2°.
• 7x: 26.2°
• 6x, 8x: 22.2° (maybe even slightly less)
• Dx: 27,2°

Pretty weird.

 

WOW...That is literally all over the place...LOL.

 

Thanks for the feedback!

[Thomas Jentzsch]

Yea, I cannot spot a pattern either. Maybe the PAL colors are mixed by two or more overlapping parameters.

 

We need at least one additional PAL setup to sort out e.g. a hardware problem on my side.

[Trebor]

Yea, I cannot spot a pattern either. Maybe the PAL colors are mixed by two or more overlapping parameters.

 

We need at least one additional PAL setup to sort out e.g. a hardware problem on my side.

Absolutely...more 2600 PAL hardware comparisons via CRT would be ideal. Also, as SeaGTGruff mentioned, what is really needed is PAL TIA schematics.

 

Until then, let's try this in the interim, attached are all the phase shifts created (including a new Stella217PAL, due to the previous note by you respecting 6x, 8x might be slightly less than 22.2 degrees): Stella2XX_Palettes_20131003.zip

 

There's a dozen palettes. Take your time (days/weeks/whatever) and when able perhaps provide a list similar to the one you provided earlier, but with all hue row values. It would be great to have one list for when the system is 'colder' (~15 minutes or less total run time) and then another for when it is 'warmer' (Over ~30 minutes or so). Regardless, even just having one list, I believe is an improvement and certainly can't hurt.

 

Once you come back with the values, I'll build it. You test it out and any updates of course can be made. Obviously we have 0x, 1x, Ex, Fx. If you can, perhaps take each palette phase shift one at a time against what you see on your television, and the best matching ones write down. Providing a list of phase shift and the associated hue row will improve at the very least emulation for one PAL user

 

Going through all the phase shifts, and writing down matches may furnish a pattern with time, you never know

 

Thanks, Thomas.

[Thomas Jentzsch]

So far I haven't noticed any color changes when the system got warmer.

[Trebor]

 

Awesome, makes it 'easier', and if there is any connection to the way NTSC is generated that definitely eliminates cross reference to NTSC Fx & Ex - values most noticed changed with the slight ~0.5 degree phase shift I see with NTSC 4-Switch Woody. The other NTSC hues the change is really not visually noticable; especially with such a modest shift.

[Trebor]

Thomas,

 

When you have a chance, please try the attached palette out, and tell me what you think/how it compares to your hardware.

 

StellaPALTEST01.zip

 

I tried a different approach, and curious how much better/worse the results are in comparison to your hardware.

 

Thanks.

[Trebor]

Going back to what you posted concerning the real hardware, I'm hoping you could please clarify something...

 

The capture here: http://atariage.com/forums/uploads/monthly_10_2013/post-45-0-07453500-1380733545.jpg

That's running Stella257.pal under the Stella emulator displayed on a computer LCD with the PAL colors arranged in more of a NTSC order, correct?

 

Then this capture: http://atariage.com/forums/uploads/monthly_10_2013/post-45-0-24277700-1380733570.jpg

That's running your PAL Vader console under a CRT with no plus or minus (hue) degrees, just the color palette from the Vader system is arranged in more of a NTSC order, correct?

 

The slight variances in the colors between the above linked captures, likely may be just the nature of the different hardware at work. Even between two LCD's/LED's, something can look green on one and cyan on another, or more orange on one and more yellow on another (Especially the cheaper and more widely used TN panels - they have some horrible differences between them and the viewing angle colors shifts are really bad).

 

The -10 degree palette, is around what is from an LCD/HDTV display type. And at -10 degree shift from the console, it would make a palette created for the setting and the PAL system in that picture showing more a simulation arrangement of a NTSC palette at ~-43 degrees. As mentioned earlier ~-45 degrees is more the LCD/Plasma/LED way an 2600 (And 7800) has their hue begin point (and overall colors) displayed (Half way between the PAL -57 degrees, and NTSC -33 degree setting).

 

From the HDTV's seen, they vary between minus ~12 to 15 degrees compared to a CRT for the Atari consoles. That's why what 'should be' more gold-ish in hue looks more yellowish-green. In order to achieve closer to typical CRT colors on an LCD/LED, you need to really turn the 'red' up. Even then it won't be just like CRT colors, but at least closer.

 

When utilizing the Stella257 palette or any other ones, tweaking the gamma, saturation, scanline intensity and whatever other settings (i.e. artifacts), can also greatly impact the colors. Using 'Custom' under TV Effects, click 'clone composite', and try 25% scanlines. Bump gamma to 75 (from the 50 default). Also a bit more saturation (Similar 50 default to 75) sometimes could be needed. CRT gamma tends to be more in the 2.35-2.55 range, while most PC monitors are @2.20, some drop as low as 1.80. Hence, the need for a bump.

 

Having more samples from actual PAL hardware to confirm what is experienced from PAL 2600 hopefully will happen in the future.

 

Anyhow, thanks again for all the input and feedback Thomas, it is greatly appreciated.

[Trebor]

Just in case there's any doubt and questioning of the screenshots used in the prior post are for the 7800 and the 2600 may not do that. Here's the color bars and Pitfall on a CRT:


Not the greatest of screen captures and it looks like Bx got caught with the refresh passing. The important point is the hue begin point, 1x, the color bar directly below the top grayscale (Second row of swatches).


Now, here's the same color bars on an LCD:

Again, not the greatest of screen captures; regardless, that is where the 'greenish-yellow' is seen in 1x - on an LCD [LED/Plasma shows similar results].

On a CRT 1x is, as many of the tech docs reference, a 'gold' or gold-ish color range.

And here is what needs to be done under a LCD to Pitfall to get it to look more like a CRT:


It isn't until the LCD is brought to ~between the minus 30 to minus 45 units (Each units looks to represent ~half a degree) setting does it look more like a CRT. Even then, it's not right because obviously everything is shifted, so that even the things that are supposed to be more green, like the tree's leaves, start to become a weak green-brown.

Ideally, it is meant for the colors to be seen and handled as a CRT would show them, not today's modern displays. They do not handle the 2600 (or 7800) color palettes correctly, no matter what adjustments you make. Sure, a greenish-yellowish 1x can be utilized under Stella or any other 2600 emulator, but it is losing the original color intent for that game on the 2600.

Summary...
PAL hue begin point = ~-57 degrees (Strong Green-yellow range)
NTSC CRT hue begin point =~-33 degrees (Gold range)
NTSC LCD/LED/Plasma hue begin point = ~-45 degrees (Weak Green-Yellow range)

Under the most optimal circumstances Stella257 should be the palette, but the 2600 can fall a little shy of a 180 degrees color burst, so Stella247 and Stella252 is provided. If a 2600 system phases even higher than ~25.7 degrees, Stella262 and Stella267 are available as well.

All palettes can be downloaded together here.

[Thomas Jentzsch]

Thomas,

 

When you have a chance, please try the attached palette out, and tell me what you think/how it compares to your hardware.

 

StellaPALTEST01.zip

PALTEST01.PNG

 

I tried a different approach, and curious how much better/worse the results are in comparison to your hardware.

 

Thanks.

I think the other palettes are better.

 

I noticed that you played with the color tint within the rows.

 

E.g. 2x looks good better from 24. But 20 and 22 have much to little red.

4x is even worse. 44 is OK, but below red is missing and above the colors gets off more and more (e.g. 4e is a bright salmon on my TV, not yellow)

6x misses a lot of red on the darker colors (e.g. 60 looks like 80 on my TV), starting from 26 is looking OK.

 

The blue colors are overall better.

 

Starting from 9x, the brighter colors have a bit too much green.

7x is way to blue on the dark colors

5x is good for the dark colors, the brighter ones have too much blue.

The same can be said for 3x. This one has no blue tint on my TV.

[Thomas Jentzsch]

Going back to what you posted concerning the real hardware, I'm hoping you could please clarify something...

 

The capture here: http://atariage.com/forums/uploads/monthly_10_2013/post-45-0-07453500-1380733545.jpg

That's running Stella257.pal under the Stella emulator displayed on a computer LCD with the PAL colors arranged in more of a NTSC order, correct?

 

Then this capture: http://atariage.com/forums/uploads/monthly_10_2013/post-45-0-24277700-1380733570.jpg

That's running your PAL Vader console under a CRT with no plus or minus (hue) degrees, just the color palette from the Vader system is arranged in more of a NTSC order, correct?

Both correct.

[Thomas Jentzsch]

Your NTSC CRT TV shot looks pretty ugly compared to my PAL TV. Besides between 3x and 2x, my color transitions are much smoother. And almost all colors look different (besides the dark greens).

[Trebor]

Your NTSC CRT TV shot looks pretty ugly compared to my PAL TV. Besides between 3x and 2x, my color transitions are much smoother. And almost all colors look different (besides the dark greens).

I'll have to take better captures...My focus was 1x. I was working with both poor lighting and not the greatest of cameras (I should have adjusted contrast, brightness, and saturation to compensate too).

 

I'm a little pressed for what I can do with my real hardware with a 2 year old running about...lol.

 

Thanks for the confirmation regarding the aforementioned captures.

 

Here's one I also took against a CRT with a section of 0x-5x 'up-close':

 

Still not the greatest, but again, I think it gives the idea of how 1x on a CRT appears different from an LCD/LED/Plasma. There is no greenish-yellow 1x on a CRT, it is in the gold range.

[Trebor]

I confirmed today that the pot adjustment for the NTSC 2600 controls the phase shift. For the NTSC 7800 the pot adjustment controls the blue to green / blue to red ratio. Logically, it makes sense why the 7800 has two separate pot controls for each system. The thing which both the 2600 pot adjustment and the pot adjustments within the 7800 share is they are highly and super sensitive. It takes a very, very, minimal turn to cause a significant change.

As mentioned earlier the phase shift of the 2600 seems to change (automatically) much less than the 7800 during consistent run-time. After an initial start-up (~2 minutes-5 minutes of running), the 2600 only changes about +0.5 to +1.0 degree, even after an hour of running. The 7800 after an initial startup can shift as much as +1.0 to +2.0 degrees in less than an hour.

Knowing that the pot adjustment changes the phase shift on an NTSC 2600 system, and presuming the PAL systems work the same way, even more the reason why the hardware differences can be a huge factor in why the PAL palettes may look slightly different across various hardware displays.

Seeing the captures from the Stella257 Palette and the 2600 Vader PAL system on a Sony TV, demonstrates the point pretty well too.

The good thing with the 2600 vs. 7800 is there seems to be no variance of the blue to green or blue to red ratios; which does eliminate some of the yet still great possible varieties.

To simulate the 2600 palette displayed on a CRT for both NTSC and PAL under Stella, these palettes serve the purpose best; while making adjustments via Stella's video options.

[Trebor]

Found a real world example that shows a different behavior than the previously mentioned two 4-switch Woody's. It demonstrates a 2600 phase shifting VERY quickly and going higher than 26.7 degrees...Around 27.7 degrees, having a similar behavior to the 7800 systems mentioned before.

 

Here's captures via Stella and one of the best and most common games - Space Invaders.

 

Notice the ground color. It starts as a strong green and works it way to a strong brown.

Captures are in this order: 24.7 --> 25.2 --> 25.7 --> 26.2 --> 26.7 --> 27.2 --> 27.7

 

 

Now here is the real hardware example. Only 54 seconds into the video, a nice green color:

 

 

Here's how the ground looks only a little more than 2 minutes later:

 

 

Here is just a little more than 5 minutes from the first capture:

 

 

See the video here: http://www.youtube.com/watch?v=tE3gGpKew6Y

 

The previous suggestions was to keep the palettes in the 24.7 thru 26.7 degrees phase shift range; but evidently, there are Atari 2600's, like the Atari 7800's, where it phase shifts even higher and changes even faster than the 7800's experienced.

 

Download the entire Atari 2600 Stella palette set with a degree phase shift from 24.7 thru 27.7 with 0.5 increments HERE.