Accurate(ish) CRT TV simulation theory

[tetsuo55]

Somewhat close. We need to scale the whole line including the left and right borders. SNES black borders must be different and this counts on the aspect ratio too. Looks like left & right borders is the part of the whole image and should be rendered on the endpoint device as well.

Exactly

[blargg]

First, refer to Figure 1, the video signal trace. You can see that the signal begins around 10 us, and ends around 63 us (us = microsecond = 1/1000000 of a second). Where it begins, the electron beam is at the left side of the screen, starting a scanline. Where it ends, the electron beam is at the right side of the screen, ending a scanline (Figure 10). Thus, it spends about 52.66 us drawing each scanline. Whatever the video signal is doing during this scanning affects what appears on that scanline.

To find what appears on a given scanline, we need to know what the video signal is doing during those 52.66 us while the electron beam scans it. What is the SNES doing during those 52.66 us? It's modulating the video signal with pixels (surprise!). The PPU generates pixels at a rate of 5369318 per second, and a scanline lasts 0.00005266 seconds, so it's generating about 5369318*0.00005266 = 282.7 pixels during the scanline time. Since it only has 256 active pixels across, it generates black for some of them, which forms a border on the left and right sides.

For another example of number of pixels across, refer to Figure 4. The part labeled Active Video is the 52.66 us section mentioned above. Scroll down to the last list of that section, where they make some calculations. That article describes a video digitizer, and they talk about how often it samples pixels each scanline. It runs at 12.15 MHz = 12150000 per second, which they calculate to capture 772 pixels per line, BUT they're including when the electron beam isn't scanning (the device captures sync information so it can process it). We can calculate the number of pixels for the ACTIVE portion as 12150000*0.00005266 = 639.8, very close to their 640.

The other dimension, number of scanlines, is specified as 486 lines (see Active Image). That's for an interlaced image, made of even and odd fields. The SNES fakes a progressive image by not alternating fields, so it only has half that many scanlines, 243. That's all there is to the vertical.

The exact DAC timings have no connection with CRT electronic impulses. With DAC timings you just know what is going on, but you can't explain how powerful the impulse is.

In ruler we trust.
[...]
But how can you calculate aspect ratio?

Width is just time units and how you are going to express SIZE for TIME?

By dividing by time per scanline: pixel width as fraction of scanline = pixel duration (time) / time per scanline

The DAC timings tell you how long each pixel lasts. If you know how long it takes to draw a scanline with the electron beam (which we do), then you can tell how wide a pixel is (% of the screen's width). If you know how many scanlines there are (which we do), and the aspect ratio of the screen (which we do, 4:3), we can calculate the relative width of a pixel as compared to the (vertical) spacing between scanlines, and thus a pixel's aspect ratio.

But a ruler would certainly bypass the intellectual chaos that is this thread.

[tetsuo55]

I see now, the scanline time in the calucation is already the image data, hsync and vsync can indeed be ignored then.

Other than that you're explenation exactly matches what i was thinking.

282.7 pixels during the scanline time. Since it only has 256 active pixels across, it generates black for some of them, which forms a border on the left and right sides.

Do we know exactly how many of these modulations are black?

If you do a straight 1.148936 conversion on the basic image the aspect ratio would still be wrong right?

[FirebrandX]

I'm dead-convinced the intended apsect ratio for the active SNES 256x224 pixels is 1.3333:1. I'm also convinced this is true of the middle 224 lines on the regular NES because I remember being able to access my old Trinitron Tv's secret menu system to shrink/expand the H/V settings and found the extra 16 pixels (8 top, 8 bottom) were cropped off on default settings. I remember being thrilled to find the "rest" of the virtical graphic for games like Metroid and Zelda.

[tetsuo55]

I'm dead-convinced the intended apsect ratio for the active SNES 256x224 pixels is 1.3333:1. I'm also convinced this is true of the middle 224 lines on the regular NES because I remember being able to access my old Trinitron Tv's secret menu system to shrink/expand the H/V settings and found the extra 16 pixels (8 top, 8 bottom) were cropped off on default settings. I remember being thrilled to find the "rest" of the virtical graphic for games like Metroid and Zelda.

Awesome

If i assume for a moment that all extra pixels rendered are black.

-Pad 256 to 282 with black pixels
-Scale with 1.148936

Crop black pixels and you have an exact aspect ratio of 13333:1

Same result with the calculation on the first page. (within a margin of error)

[blargg]

282.7 pixels during the scanline time. Since it only has 256 active pixels across, it generates black for some of them, which forms a border on the left and right sides.

Do we know exactly how many of these modulations are black?

Since we know how many SNES pixels are within the active region, and how many SNES pixels are non-black, we can determine how many are black by subtraction: 282.7-256=26.7 black.

If you do a straight 1.148936 conversion on the basic image the aspect ratio would still be wrong right?

If you take a 256x224 SNES image and display it with pixels that are 1.148936 times as wide as they are tall, the image will be correct. Nothing else matters.

I'm dead-convinced the intended apsect ratio for the active SNES 256x224 pixels is 1.3333:1

I thought we were trying to find what it IS, rather than what its designers intended. In any case, as I covered in another post, if we displayed a 256x224 image completely filling a 4:3 frame, the aspect ratio of the INDIVIDUAL PIXELS would be 1.166667:1, which is pretty close to 1.148936:1 (within 1.5% of each other). Are you quibbling over the 1.5% difference between 1.166667 and 1.148936, or do you think each individual pixel should be 1.3333 times as wide as it is tall?

I'm also convinced this is true of the middle 224 lines on the regular NES because I remember being able to access my old Trinitron Tv's secret menu system to shrink/expand the H/V settings and found the extra 16 pixels (8 top, 8 bottom) were cropped off on default settings. I remember being thrilled to find the "rest" of the virtical graphic for games like Metroid and Zelda.

That's probably one reason they added borders on the top and bottom of the SNES frame: they're cut off on most TVs, and blanking them gives more vblank time for games to do graphic operations each frame.

[_willow_]

282.7

I do really doubt the fractional timing. There is something else giving the fractional tail, maybe ray back retrace. It must be expressed with the same time units as the visible image but takes different time on the real tube.

[FitzRoy]

Firebrand, blargg is right. There are two equally arguable "corrects." One concerns itself with how the art was intended to look, the other concerns itself with what shitty technology at the time actually did to it. Obviously, I see no reason why I should be using the latter now that I have a choice, and perhaps you feel the same.

[Dullaron]

How I get bsnes screen to show as 5:4 instead of 4:3? Does it auto set it?

[FirebrandX]

Firebrand, blargg is right. There are two equally arguable "corrects." One concerns itself with how the art was intended to look, the other concerns itself with what shitty technology at the time actually did to it. Obviously, I see no reason why I should be using the latter now that I have a choice, and perhaps you feel the same.

Agreed. Shitty technology aside, I'm going with what the artists' intended aspect of 1.3333:1 for the 256x224 area.

Getting back to my old Trinitron TV, I remember setting the H and V size & position to exactly match the SNES 256x224 area. This combined with the "hot" new technology of the S-Video cable for the SNES gave me the best looking picture in town at the time. People were like "Damn that's a good picture!" and I would show them the S-Video cable so they could buy one themselves.

[Gil_Hamilton]

Firebrand, blargg is right. There are two equally arguable "corrects." One concerns itself with how the art was intended to look, the other concerns itself with what shitty technology at the time actually did to it. Obviously, I see no reason why I should be using the latter now that I have a choice, and perhaps you feel the same.

Agreed. Shitty technology aside, I'm going with what the artists' intended aspect of 1.3333:1 for the 256x224 area.

Getting back to my old Trinitron TV, I remember setting the H and V size & position to exactly match the SNES 256x224 area. This combined with the "hot" new technology of the S-Video cable for the SNES gave me the best looking picture in town at the time. People were like "Damn that's a good picture!" and I would show them the S-Video cable so they could buy one themselves.

And I was forced to a composite monitor from 1982. With MONO SOUND!

I remember railing against the s video modes in ZSNES back in the day, and telling people to mess with the knobs on their monitors instead.

[Verdauga Greeneyes]

Firebrand, blargg is right. There are two equally arguable "corrects." One concerns itself with how the art was intended to look, the other concerns itself with what shitty technology at the time actually did to it. Obviously, I see no reason why I should be using the latter now that I have a choice, and perhaps you feel the same.

Of course, it does depend on the amount of testing developers did. If they wanted their image to look truly right on the average TV, they would not have gone with the 4:3 aspect ratio but adjusted it slightly (1.5%). It's possible that some games were intended to use whichever aspect ratio the devs observed, which would be closer to our 1.148936:1 one, and some games were intended to use 1.166667:1...

Anyway, thank you blargg for being so patient with this topic.

Edit: by the way, are there any NTSC ports of PAL games which look peculiar at the NTSC aspect ratio? Our aspect ratio deviates from 4:3 a lot more than yours does, which makes discussing it more interesting in my opinion.

[FirebrandX]

Of course, it does depend on the amount of testing developers did. If they wanted their image to look truly right on the average TV, they would not have gone with the 4:3 aspect ratio but adjusted it slightly (1.5%). It's possible that some games were intended to use whichever aspect ratio the devs observed, which would be closer to our 1.148936:1 one, and some games were intended to use 1.166667:1...

This would not explain why geometric shapes like circles and triangles only become "perfect" when the 1.3333:1 aspect ratio is applied to the NTSC SNES 256x224 graphic area. I've compared circles, triangles, and squares on several games that render things like planet discs, and these shapes become perfectly geometric ONLY when I apply the 4:3 correction. Why is that? The only answer is this is what they intended when they designed the games in the first place.

[Verdauga Greeneyes]

This would not explain why geometric shapes like circles and triangles only become "perfect" when the 1.3333:1 aspect ratio is applied to the NTSC SNES 256x224 graphic area. I've compared circles, triangles, and squares on several games that render things like planet discs, and these shapes become perfectly geometric ONLY when I apply the 4:3 correction. Why is that? The only answer is this is what they intended when they designed the games in the first place.

Well, those games were obviously intended to be 4:3. I'm just saying it's a stretch to assume all developers did this.

[blackmyst]

Firebrand, blargg is right. There are two equally arguable "corrects." One concerns itself with how the art was intended to look, the other concerns itself with what shitty technology at the time actually did to it. Obviously, I see no reason why I should be using the latter now that I have a choice, and perhaps you feel the same.

Ok ok wait. Aren't those pretty much one and the same? "How it was intended to look" is exactly "here's your target display, that's what you're pixeling for".

I hope you're not advocating to turn off all filters and relinquish an actual unified image in favour of staring at a bunch of very sharp stacked squares.

Maybe I'm reading into this wrong?

[FitzRoy]

Firebrand, blargg is right. There are two equally arguable "corrects." One concerns itself with how the art was intended to look, the other concerns itself with what shitty technology at the time actually did to it. Obviously, I see no reason why I should be using the latter now that I have a choice, and perhaps you feel the same.

Ok ok wait. Aren't those pretty much one and the same? "How it was intended to look" is exactly "here's your target display, that's what you're pixeling for".

The artists were probably told that the end ratio would be the standard 1.33:1 not 1.32:1 or whatever deviation you guys took the time to meticulously calculate in your basements. There isn't a chance in hell any of the artists were told to draw their art 1% wider, as if there would be any benefit to telling them this bullshit because (a) it wouldn't really be noticeable and (b) it's not even possible to take such a slight corrective action over 4/3 at such a low resolution. Furthermore, when their art was ported to PAL systems which output to completely different tvs which stretched it more, it doesn't automatically mean that the artist simulataneously intended two different results simply because that's what happened. These are logical assertions supported by evidence.

Now, we know that there are games which ignored compensation completely or in certain scenes. And there probably were some PAL only games that were drawn specifically for PAL tv behavior.

I think blargg is right for the defaults, I simply want an easier way than dual resolutions with dual correction settings in advanced that don't account for NTSC overscan.

I hope you're not advocating to turn off all filters and relinquish an actual unified image in favour of staring at a bunch of very sharp stacked squares.

Who said anything about 8:7 defaults or removing filters?

[blackmyst]

The artists were probably told that the end ratio would be the standard 1.33:1 not 1.32:1 or whatever deviation you guys took the time to meticulously calculate in your basements.

Oh god please don't drop me in with tetsuo55 or whoever else. I'm not trying to calculate aspect ratios to the 100th decimal. Such a tiny difference is hardly visible.

I'm responding to your idea about the "difference" between what was -intended- and what we actually saw. What makes you think developers didn't have any SNES kits hooked up to TV's? I doubt artists were "told" anything, unless it was to just see for themselves what their pixel art was gonna end up looking like on a devkit.

it's not even possible to take such a slight corrective action over 4/3 at such a low resolution.

Correction? No, but it's definitely gonna be different when building a whole screen of pixels from scratch. Like I said, not very significant, unlike the difference between the near-gameboy ratio you get with square pixels, and 4:3 (are there still people who support the former?). But still. Semantics etc.

Who said anything about 8:7 defaults or removing filters?

I said I might be reading into this wrong, but from your comment I got the impression you believe people were developing for something else entirely than the actual TV's their graphics were gonna end up on. Hence my reply.

[blargg]

This would not explain why geometric shapes like circles and triangles only become "perfect" when the 1.3333:1 aspect ratio is applied to the NTSC SNES 256x224 graphic area. I've compared circles, triangles, and squares on several games that render things like planet discs, and these shapes become perfectly geometric ONLY when I apply the 4:3 correction. Why is that?

First off, I just want to say that I don't care about a 1.5% difference, given that TVs themselves vary by more than this. Developer intention, whatever, it's miniscule. So I'm not arguing either way regarding developer intention. Only each developer knows what his intention was, and we don't have any present, so it's just speculation. Speculation is fine as long as reasons are offered, rather than personal attacks.

I haven't done the math yet, but it might have been simpler to scale things horizontally by 1.1666667 than by 1.148936. Perhaps they drew the graphics inside 320x240 canvases, then contracted them to 256x240 or something. Such scaled shapes might also have more fully used some group of 8x8 graphic tiles. Given how close it is, it is conceptually simpler to treat the 256x224 area as fitting within the TV.

Hardware-wise, the SNES pixel clock is related to the colorburst frequency (colorburst*1.5) and scanline height is fixed by NTSC, so the only thing that could be easily fine-tuned to get an exact 4:3 image aspect ratio is the width or height. Adjusting width to something other than 256 would have made software a lot more tedious. A height of 220.6 would have given a 4:3 image aspect ratio, and 224 is the closest multiple of 8 to that.

Anyway, I thought the goal of all this was to recreate what actually appeared on TVs, rather than what developers intended or what appeared on one's custom-adjusted TV (I know I have always adjusted my TVs to display more overscan for video games, heh). In other words, recreate what we saw when we played the games back in the day. Of course, back in the day we wanted RGB displays with overscan, but now nostalgia is usually best served by what we actually had.

The artists were probably told that the end ratio would be the standard 1.33:1 not 1.32:1 or whatever deviation you guys took the time to meticulously calculate in your basements.

You obviously feel very strongly about this issue. Is it simply because you want to be sure emulators don't remove the option for the image to be displayed with your preferred ratio? I'm sure most can, and even if they can only show it as hardware does, 1.5% narrower horizontally, I doubt you'd even notice unless one of us maticulously calculated it in our basement.

[FitzRoy]

Someone told a newcomer who thought the 4:3 standard was the correct ratio that it wasn't the correct SNES ratio. I'm purely against people equating the televisions with developer intent and telling people that it's somehow less accurate to use 4:3 based on false assumptions connecting display technology to intent on every level despite logic and evidence to the contrary. Let's review:

-Glitches were not intended, yet they occurred. Occurrence = intent argument broken, and easily.
-PAL ports of NTSC games never had their art changed to compensate for the different correction. Simultaneous intent is impossible. This also resulted in undesirable and noticeable over-stretching of practically every NTSC port when played on a PAL television.
-Evidence has been found showing developers drawing for 8:7, resulting in ovals rather than circles on worldy objects known to be circular. Seem likely that a 1% deviance from 4:3 was given any concern when complete disregard for ANY correction was rampant?
-Evidence has been found showing known equilateral shapes achieving less equilateral proportions under the 1% deviance than 4:3 itself.
-4:3 was the physical aspect ratio of the television, and THE standard by which all material of this era is was designed.
-corrective forethought for pixel artists on a resolution this low almost certainly wouldn't have been possible to fine tune to 7/47ths. 1/6th is far more likely.

That's all I'm saying. I just don't want things to get so pedantic over here that we're shooting down everyone who thinks its simply 4:3.

[Gil_Hamilton]

Someone should set up a variable bank, so you can define YOUR SPECIFIC TELEVISION's overscan and stretch values.
Since, you know, every single TV was different, and very few were properly calibrated.

That'd lay this argument to rest. Then we'd just have to worry about accurate CRT*backspace*DVD recorder simulation filters.

[blargg]

I just don't want things to get so pedantic over here that we're shooting down everyone who thinks its simply 4:3.

Do you have any problem with the following?

The NTSC SNES image aspect ratio is approximately 4:3, and some developers probably had their artists treat it as exactly 4:3.

The NTSC SNES pixel aspect ratio is NOT 4:3.

[tetsuo55]

okay i have one last question.

I think we can all agree that the CRT-TV's the snes was designed for was completely unable to display a progressive signal.

although the snes internally computes a certain height the full signal will eventually be 525 lines high. Unlike interlaced both frames simply display the same image.(effictively creating a 30 fps progressive image on a 60fps screen by line doubling)

IF this is right the image is not 282,7x243 but 282,7x486

486 means 648 pixels wide

648/282,7 = 2.2921825256455606650159179342059

--------
I think this is right because:
-Tv's only accept a ntsc or pal broadcast signal
-The calculated aspect ratio cannot be right for both progressive and interlaced
-I cannot find anywhere that the thickness of a scanline can be adjusted by changing the signal (unless you use a multisync monitor which changes the vertical refreshrate)

Someone should set up a variable bank, so you can define YOUR SPECIFIC TELEVISION's overscan and stretch values.
Since, you know, every single TV was different, and very few were properly calibrated.

That'd lay this argument to rest. Then we'd just have to worry about accurate CRT*backspace*DVD recorder simulation filters.

After my question above i can continue with the CRT-TV simulation.
From what i've read the "Standard" was to have -5 to +5 % overscan options in the secret menu. Not every TV did this but i think its a good idea to add this to the CRT-TV simulator, then everyone can tweak the screen to the aspect ratio they want.

I used "" around the word standard because the word had no meaning in tv land

[Gil_Hamilton]

okay i have one last question.

I think we can all agree that the CRT-TV's the snes was designed for was completely unable to display a progressive signal.

although the snes internally computes a certain height the full signal will eventually be 525 lines high. Unlike interlaced both frames simply display the same image.(effictively creating a 30 fps progressive image on a 60fps screen by line doubling)

IF this is right the image is not 282,7x243 but 282,7x486

486 means 648 pixels wide

648/282,7 = 2.2921825256455606650159179342059

--------
I think this is right because:
-Tv's only accept a ntsc or pal broadcast signal
-The calculated aspect ratio cannot be right for both progressive and interlaced
-I cannot find anywhere that the thickness of a scanline can be adjusted by changing the signal (unless you use a multisync monitor which changes the vertical refreshrate)

Except that scan position is controlled by the system.
As I understand things, ye olde systems muck with the timing signal to get the same set of scanlines drawn twice. It confuses the hell out of modern panel TVs, but works GREAT on CRTs.

And you may notice that the dark scanlines disappear(or at least shrink) when the system is in "high-res" mode. Because it's now drawing picture inside the black lines.

[tetsuo55]

Except that scan position is controlled by the system.
As I understand things, ye olde systems muck with the timing signal to get the same set of scanlines drawn twice. It confuses the hell out of modern panel TVs, but works GREAT on CRTs.

And you may notice that the dark scanlines disappear(or at least shrink) when the system is in "high-res" mode. Because it's now drawing picture inside the black lines.

The result is the same, the image is twice as high

the question is how does it work exactly?

[creaothceann]

I think we can all agree that the CRT-TV's the snes was designed for was completely unable to display a progressive signal.

although the snes internally computes a certain height the full signal will eventually be 525 lines high. Unlike interlaced both frames simply display the same image.(effictively creating a 30 fps progressive image on a 60fps screen by line doubling)

Let's recap:

The TV draws 525 lines interlaced in two passes at 60 fields per second / 30 frames per second. The signal source is responsible for shifting every other field.

The SNES doesn't shift the fields in "progressive mode", so we get a 60 Hz progressive image with 525 / 2 visible lines, and black scanlines between them. In "interlaced mode" the SNES includes the shifting signals, so we get a 30 Hz interlaced image with 525 visible lines and no (or less pronounced) scanlines.

Aspect ratio can be safely assumed to be 4:3. Let's drop that already.