Can someone EXPLAIN the whole "720x480" thing to me?

[CursedLemon]

Now wait. I think through all this I don't ever actually recall understanding why non-square pixels are used at all.

[edDV]

Now wait. I think through all this I don't ever actually recall understanding why non-square pixels are used at all.

A few reasons for standard definition.

- 704x480 (720x480 with pads) dates back to CCIR-601 (1985) for digital broadcast (aka D1). The 13.5MHz horizontal sample rate was chosen for compatibility for both NTSC and PAL. That allowed common NTSC/PAL equipment and eliminated horizontal resampling for NTSC to PAL conversion. Analog NTSC had 486 active lines. This was reduced to 480 lines for memory efficiency. Thus a 4x3 square pixel frame became 640x480. Since PAL has 576 lines, a square pixel SD PAL frame became 768x576.

- CCIR-601 was extended in the late 80's to allow 4:3 and 16:9 programming using the same transmission and recording equipment. The data stream is similar except for the wide flag. Display equipment looks for the flag so it knows to display as 4:3 or 16:9.

- In the early 90's MPeg2 and later DVD adopted the same resolutions and sample rates as CCIR-601 broadcast. DVD added half D1 (352x480 and 352x576), MPeg2 scaled down from there to 352x240, 352x288 and lower for internet and telecom video.

- In the late 90's ATSC and DVB digital broadcasting also used the CCIR-601 (ITU-Rec-601) spec for standard definition. Later VC1 and h.264 AVC adopted the same resolutions.

All this was done by the broadcast and telecommunications industry. It wasn't until ~Y2K that the computer industry looked up from their 1MByte RGB square pixel frame buffers.

[edDV]

HD broadcasting could have gone to non-square pixels for transmission efficiency. For example few could tell the difference between 1440x1080 and 1920x1080 with current equipment. 1440x1080 uses two thirds the bit rate for equivalent compression or would show higher quality at the same bit rate. Consumer video quality is limited more by compression formats than resolution. Still the standards bodies went with 1920x1080 to allow headroom for quality improvement under the assumption that compression quality will continue to improve.

A major accomplishment for the HD standards committees has been world wide agreement on square pixel 1920x1080 and 1280x720 high definition resolutions. Frame rate disparity still exists between "NTSC" and "PAL regions due to the need for standard definition interoperability.

The international film and TV program producers are converging on 24p (23.976p) as an international media exchange standard for "movies" and low motion TV drama.

[CursedLemon]

Boy, this feels like all of my audio education hitting me like a sack of potatoes all over again. Thanks for being so informative.

[2Bdecided]

@edDV,

There are plenty of anamorphic HD broadcasts (e.g. BBC HD is 1440x1080 in the UK, and AFAIK several in the USA are 1440x1080 or 960x720, depending on the platform) and anamorphic HD recording formats (HDCAM (not SR), DVC-Pro HD, and of course HDV).

Cheers,
David.

[2Bdecided]

- 704x480 (720x480 with pads) dates back to CCIR-601 (1985) for digital broadcast (aka D1). ... Analog NTSC had 486 active lines. This was reduced to 480 lines for memory efficiency.

That wasn't spec 601. It will happily let you digitise 486 lines - in fact you can have the whole VBI (excluding syncs) if you want!

Rec. ITU-R BT.656-4, which defines SDI (and an obsolete parallel interface) i.e. ways of transporting D1 video - does define the vertical blanking - but still includes 488 lines for NTSC. (It includes the CGMS-A and CC lines. I think previous issues of the spec missed the CGMS-A lines, making it 486 exactly).

I think it's only the emergence of MPEG (DVD, digital broadcasting) which brought us to 480-lines for NTSC video - though PCs were there much earlier. It's a multiple of 16, which works well for both.

Much simpler for PAL - it's always been 575 analogue, 576 digital.

Cheers,
David.

[lordsmurf]

I don't know that I'll ever find use for this information. Interesting, however.

[totalz]

What about a NTSC DVD with DAR of 2.35:1? Is it still encoded in 720x480? In which case the display resolution would be 1128x480? And the PAR would be 47:20?

[jagabo]

There are no DVDs with a DAR of 2.35:1. 2.35:1 movies are letterboxed in 16:9 DAR 720x480 frames.

[poisondeathray]

What about a NTSC DVD with DAR of 2.35:1? Is it still encoded in 720x480? In which case the display resolution would be 1128x480? And the PAR would be 47:20?

It still has a frame of 720x480, but there are black borders encoded into the top and bottom (letterboxing). The PAR is the same 32:27 (or 40:33 if using MPEG4 specification). The frame will display as 852x480 (or 853x480) in most media players, but there will be letterboxing

whoops jagabo too quick

[jagabo]

Sometimes 2.35:1 movies are letterboxed into 720x480 4:3 DAR frames. Yuck!

[edDV]

@edDV,

There are plenty of anamorphic HD broadcasts (e.g. BBC HD is 1440x1080 in the UK, and AFAIK several in the USA are 1440x1080 or 960x720, depending on the platform) and anamorphic HD recording formats (HDCAM (not SR), DVC-Pro HD, and of course HDV).

Cheers,
David.

ATSC currently is square pixel MPeg2 for HD. I considered going into QAM and DVB but that would only confuse the OP who is in North America. I explained the production formats above but for ATSC HD broadcast square pixels rule.

[edDV]

- 704x480 (720x480 with pads) dates back to CCIR-601 (1985) for digital broadcast (aka D1). ... Analog NTSC had 486 active lines. This was reduced to 480 lines for memory efficiency.

That wasn't spec 601. It will happily let you digitise 486 lines - in fact you can have the whole VBI (excluding syncs) if you want!

Rec. ITU-R BT.656-4, which defines SDI (and an obsolete parallel interface) i.e. ways of transporting D1 video - does define the vertical blanking - but still includes 488 lines for NTSC. (It includes the CGMS-A and CC lines. I think previous issues of the spec missed the CGMS-A lines, making it 486 exactly).

I think it's only the emergence of MPEG (DVD, digital broadcasting) which brought us to 480-lines for NTSC video - though PCs were there much earlier. It's a multiple of 16, which works well for both.

Much simpler for PAL - it's always been 575 analogue, 576 digital.

Cheers,
David.

I was trying to avoid details. Analog NTSC was 486 lines plus 2 vertical interval. There was a big spat in the mid 80's whether broadcast of 480 lines with 3 line black crops top and bottom violated FCC rules. Computer frame buffers of the day were 640x480 and the ram cost to go 486 lines was costly. This affected most character generators, weather machines, frame syncs and CGI full frame effects. The second issue related to A/D of analog video. It was argued previous production required 486 lines because of creative intent of the director. The digital TV geeks claimed a slight letterbox crop was in the overscan area and should be allowed.

The FCC studied the issue for many many months causing much pain for us digital equipment makers. Finally they wimped out saying letterbox was a programming decision. 480 line letterbox with 6 fill lines was in program space and thus not regulated. Later the standards were amended to allow 480 lines.

I previously posted a link detailing even more digital video vs. legacy controversy. I'll try to find it.

[edDV]

PS: Here it is and it has some updates. Proves that everything we know is wrong.
http://lipas.uwasa.fi/~f76998/video/conversion/

More useful reading
http://www.bbc.co.uk/commissioning/tvbranding/picturesize.shtml

[totalz]

Thanks jagabo & poisondeathray for the info...

This may be off-topic, but if I want to transcode a 16:9 720x480 NTSC dvd, do I keep the same setting?
Or 854*480? I prefer quality

[Alex_ander]

if I want to transcode a 16:9 720x480 NTSC dvd, do I keep the same setting?
Or 854*480? I prefer quality

If you want to transcode to something 'square pixel', then 852x480 (mod4) is a possible final size. However by ITU, in original video only 704x480 samples of those 720x480 represent the 16:9 AR part of video, so for correct proportions you need to crop to those numbers just before resizing.
The visible part of image within 720x480 may be wider than 704 (like 711x480 for NTSC DV/DVD). If you don't want to crop off any visible bit, here's another option: first resize 720x480 to 872x480, then crop horizontally to visible (rounding to mod4). The 872 number comes from 853x720/704 proportional compensation.

Apart from AR considerations, take care of interlacing, telecine or whatever type of the original video.

[totalz]

if I want to transcode a 16:9 720x480 NTSC dvd, do I keep the same setting?
Or 854*480? I prefer quality

If you want to transcode to something 'square pixel', then 852x480 (mod4) is a possible final size. However by ITU, in original video only 704x480 samples of those 720x480 represent the 16:9 AR part of video, so for correct proportions you need to crop to those numbers just before resizing.
The visible part of image within 720x480 may be wider than 704 (like 711x480 for NTSC DV/DVD). If you don't want to crop off any visible bit, here's another option: first resize 720x480 to 872x480, then crop horizontally to visible (rounding to mod4). The 872 number comes from 853x720/704 proportional compensation.

Apart from AR considerations, take care of interlacing, telecine or whatever type of the original video.

Thanks for the mod4 reminder : )
ITU - International Telecommunication Union??? Probably not!?
Googled it : Results 1 - 10 of about 36,600,000 for ITU [definition]. (0.16 seconds)

Does it improve quality with square pixel? I mean 1:1 is always good, but since the source is not, does square pixel works better with H264?

How can I find out the actual image width, in the case as you've mentioned above?

[Alex_ander]

ITU - International Telecommunication Union?

Yes, that's from its standards.
http://www.itu.int/en/pages/default.aspx

Does it improve quality with square pixel?

Only in case your resize filter is better than that of a playback device. The stretched video will need higher bitrate for the same encoding quality. But it is compatible with devices which don't support AR flags. Otherwise you can use a container like matroska which supports those flags, and use original resolution.

How can I find out the actual image width, in the case as you've mentioned above?

Actual image width depends on how the video was produced. In theory, it is about 711 just after capture but it may have been cropped to 704 at production. If 711, then the visible part will be 711x872/720=860, not exactly 16:9 of course (same as in source).

[2Bdecided]

PS: Here it is and it has some updates. Proves that everything we know is wrong.
http://lipas.uwasa.fi/~f76998/video/conversion/

No, that page is wrong. He assume 576 lines for analogue PAL height. It was 575. Which means most of the "problems" he cites become less than one pixel!

http://forum.doom9.org/showthread.php?p=1072530#post1072530
http://forum.doom9.org/showthread.php?p=1100187#post1100187

More useful reading
http://www.bbc.co.uk/commissioning/tvbranding/picturesize.shtml

It's not that useful. We know a "perfect" analogue PAL signal is exactly 702x575 in D1. They just invent a pixel aspect ratio to give 702x576 instead, and work from that. 704x576 is more accurate and more useful.

Angels on a pin head anyway - most production treats 720x576 as defining the active picture now - an error of 2.3% (16 pixels) - it makes these single pixel errors kind of irrelevant.

Cheers,
David.

[totalz]

[quote="Alex_ander"]

... container like matroska which supports those flags, and use original resolution.

How can I find out the actual image width, in the case as you've mentioned above?

Actual image width depends on how the video was produced. In theory, it is about 711 just after capture but it may have been cropped to 704 at production. If 711, then the visible part will be 711x872/720=860, not exactly 16:9 of course (same as in source).

Great, I guess I will stick to the concept of anamorphic, less bitrate, smaller file size : )

According to wiki, http://en.wikipedia.org/wiki/Dvd_video. A full D1 NTSC is 720x480, so, is there such a thing as 720 visible width? Cause I found the PAR for NTSC 16:9 is 40:33, which is perfect for 704x480! So, is there a software I can use to determine the actual visible width? Or I will have to do as you suggest "first resize 720x480 to 872x480"?

[Alex_ander]

According to wiki, http://en.wikipedia.org/wiki/Dvd_video. A full D1 NTSC is 720x480, so, is there such a thing as 720 visible width? Cause I found the PAR for NTSC 16:9 is 40:33, which is perfect for 704x480!

PAR can't be a reason for anything (it can be a result of smth. at best). The things like PAR are artificially derived from real things, defined in standards. The standard method of conversion analog NTSC to digital (720 samples per TV line, 13,5 MHz sample frequency) leads to such a position of samples in time, that ~711 of 720 belong to visible part of analog TV line, the rest are in adjacent analog blanking intervals. Those 711 correspond to horizontal dimension of a 4:3 (16:9) analog image. Vertically there are 485 analog lines defining the second visible dimension of that image. Since only 480 lines (= vertical samples) are used (for rounding numbers to mod 16 etc.) and the rest are dropped, the 711x480 can't be 4:3 (16:9) while 704 can, so that number was adopted (and both are mod16) and that is why we have 704x480 samples responsible for the declared AR's. Combining those numbers with displayed AR, one can get those 'PAR's' (if desired).
The rest of 711 are present and visible within 720 (unless intentionally blanked at further digital processing) but they don't define the nominal AR.

So, is there a software I can use to determine the actual visible width? Or I will have to do as you suggest "first resize 720x480 to 872x480"?

You can use a cropping tool in AVSP+AviSynth or Vdub and look at the crop numbers while watching the screen, or use some screen measuring tools.

[jagabo]

http://lipas.uwasa.fi/~f76998/video/conversion/ -- that page is wrong. He assume 576 lines for analogue PAL height. It was 575.

Where are you getting 575 lines? Yes, the top and bottom scanlines are half scanlines, but they are half width, not half height. If you look at the analog signal on a scope the top scanline starts half way across the width of the picture, the bottom scanline ends halfway across the picture. If you capture those lines you capture a full line of pixels for each, 576 lines total. If you underscan the picture on a CRT you'll see 576 scan lines with only the right half with picture information on the top scanline and only the left half on the bottom scanline.

In a digitized NTSC signal the active 4:3 picture with is ~711 pixels and the active signal height is 486 pixels. Since we throw away 6 scanlines from the height we must throw away some pixels from the width. 711 * 480 / 486 ~= 702. But we use 704 to keep a mod16 frame size.

[Alex_ander]

If you underscan the picture on a CRT you'll see 576 scan lines with only the right half with picture information on the top scanline and only the left half on the bottom scanline.

ITU-R BT.1700 (2005, Russian version)
(table 4)
The number of active lines---------------------------- 575.

The basic analog AR can't be defined for a different number of lines. If you count analog lines in only left or right part of the screen, it will be 575 (without the second half-line in the opposite part).

In case of DV camera it is possible to capture 576 full digital lines due to 1 line shift in DV standard from ITU for PAL and narrower analog field blanking. But image proportions still come from analog line count and if you use 576 to define AR in digital form, the horizontal number of samples will be also different from analog line length (which is 702).

[2Bdecided]

http://lipas.uwasa.fi/~f76998/video/conversion/ -- that page is wrong. He assume 576 lines for analogue PAL height. It was 575.

Where are you getting 575 lines? Yes, the top and bottom scanlines are half scanlines, but they are half width, not half height.

Ah, but they are. On a CRT at least.

They look like this:
http://www.pembers.freeserve.co.uk/World-TV-Standards/Raster-Top.jpg
from here:
http://www.pembers.freeserve.co.uk/World-TV-Standards/Line-Standards.html

In an analogue video signal, each sync pulse is there to drive one or other circuit in a TV to ensure the vertical scan (50Hz) and horizontal scan (15.625kHz) are synced to the incoming signal, such that the picture ends up in the right place, with the correct field alignment and interlacing. The circuits in conventional TVs have very little "intelligence" - it's the analogue video signal itself that defines these things.

And that signal drives those circuits to give lines where each line's height is proportional to its length - and in a very real sense, a picture with a half line at the top and the bottom is shorter (or longer!) than a picture with a full line at the top and bottom, and also closer to being "square" than one with a full line at the top and bottom.

Don't forget: they could easily have not had those half lines in the analogue standard - but they chose to put them there. Why? Well, there are lots of things they help with post-fact - but one undeniable feature is that they make the picture a rectangle, rather than a parallelogram.


Now, the idea that any of this really mattered (down to the equivalent of one modern pixel) in the analogue world is a bit fanciful - but the system is built properly to have a theoretical "perfect" alignment and adjustment and signal timing that gives a picture, in digital parlance, that's exactly 702x575.


Does it matter? No. Especially not now everything is digital.

But that page by Jukka Aho takes great delight in pointing out this "dirty little secret". Whereas, as long as you excuse people for rounding to the nearest pixel (and what else do you expect them to do?!) there is no dirty little secret. Just a fundamental misunderstanding on that page, which created an apparent pixel or two of error that doesn't really exist!

So the shocking conclusion is that most of the international standards are correct, and do work.

Cheers,
David.

[jagabo]

Ah, but they are (half height). On a CRT at least.

They look like this:
http://www.pembers.freeserve.co.uk/World-TV-Standards/Raster-Top.jpg

OK, I understand where this argument comes from now. Given a rectangular 4:3 display the lines are displayed slanted (a simple linear ramp is used for the vertical position of the electron beam) such that the top and bottom lines are half outside the display. Hence they are "half height" on the physical display resulting in a total of 575 lines.

But by this same argument, the individual scanlines are slightly longer than the 4 part width of the physical display. And the height of that slanted image (along its own, slanted, axis, not the axis of the physical screen) is slightly taller than the 3 part height of the physical display.

In any case, we all agree that these small details aren't worth worrying about. With digital video files you treat 704x480 and 704x576 as the 4:3 or 16:9 image. The extra 16 pixels used to fill a 720 pixel wide frame may or may not be displayed by any particular device but don't count as part of the 4:3 or 16:9 image.

[edDV]

More useful reading
http://www.bbc.co.uk/commissioning/tvbranding/picturesize.shtml

It's not that useful. We know a "perfect" analogue PAL signal is exactly 702x575 in D1. They just invent a pixel aspect ratio to give 702x576 instead, and work from that. 704x576 is more accurate and more useful.

Angels on a pin head anyway - most production treats 720x576 as defining the active picture now - an error of 2.3% (16 pixels) - it makes these single pixel errors kind of irrelevant.

Cheers,
David.

So long as 704x480/576 is 16x9 and the extra pixels represent additional picture width. This is necessary for proper upscale/downscale to/from 1920x1080 or 1280x720. In a production context all players need to follow common rules or graphics, mattes and object layers fail to line up in the composite forcing a rescale or worse a recode for compressed layers.

These issues cause work rejection and lost work hours. I refuse to start a project unless picture standards are agreed in the contract.

If the client wants 720x480 to be 16x9 he needs to say so up front and understand the issues. For example, 720x480 gets cropped to 704x480 for ATSC or cable SD broadcast and/or the sides will be cut off in a 1080i or 720p upscale.

[2Bdecided]

But by this same argument, the individual scanlines are slightly longer than the 4 part width of the physical display. And the height of that slanted image (along its own, slanted, axis, not the axis of the physical screen) is slightly taller than the 3 part height of the physical display.

No, that doesn't work. As displayed, they're 4 by 3 and square. The vertical edges really are vertical. If you line them up so the slanted "horizontal" lines are dead straight (perfectly horizontal) then the whole thing is skewed.

I'm all for saying that 704x576 and 704x480 are the correct aspect ratio. That's what I've been saying since I worked through it last year.

It's that website, claiming a dirty little secret, that says otherwise. And the point is, that's wrong - 704x576 and 704x480 really are correct to the nearest pixel, so they're both convenient and accurate.

Shame the world is now often using 720 instead!

Cheers,
David.

[jagabo]

But by this same argument, the individual scanlines are slightly longer than the 4 part width of the physical display. And the height of that slanted image (along its own, slanted, axis, not the axis of the physical screen) is slightly taller than the 3 part height of the physical display.

No, that doesn't work. As displayed, they're 4 by 3 and square. The vertical edges really are vertical.

But the lines are scanned diagonally -- so they are longer than the width of the 4:3 screen. Although, given the ~0.1 degree angle, the difference is far, far less than a pixel. (I made a mistake in my first back-of-the-napkin calculation and thought it was about 1 pixel but it's closer to 1/1000 of a pixel.) So I'll have to agree, the top and bottom scanlines can be considered half height under this scenario.

[manono]

Shame the world is now often using 720 instead!

Not always. Classic 'fullscreen' films, for example, are 1.37:1 rather than 1.33:1, and using the full 720x480/576 on DVD is closer to being accurate.

http://en.wikipedia.org/wiki/Academy_ratio

[edDV]

Shame the world is now often using 720 instead!

Not always. Classic 'fullscreen' films, for example, are 1.37:1 rather than 1.33:1, and using the full 720x480/576 on DVD is closer to being accurate.

http://en.wikipedia.org/wiki/Academy_ratio

That is the usual argurement that 720 doesn't matter for DVD but it means a certain level of incompatibility with digital broadcasting or most telecommunications which chop to 704 before scaling.

I don't know for sure but I'd bet most display scalers do the same.