PAL to NTSC using pulldown excessive judder

[jagabo]

There's no point in wasting time moving the electron beam back to the left after each scan line, and back to the top after scanning the entire field, so each is done as quickly as possible.

Um, it does move back to the left after each scan line, and back to the top after scanning the entire field, doesn't it?

Yes, of course. I meant it is done much more quickly than the forward scanning.

[Skiller]

Wow, mistateo you really did go down the rabbit hole. It took me years to ingest and process the information on the pages you linked to (and do tests on my own to see the actual effects).

In any case, to summarize, my source is what it is, round looks round, and after straight resize from 720x576 to 720x480 round still looks round on my (digital) TV, whether cropping to 704 or keeping the original 720.

Does this mean that both, 720 and cropped to 704, look identical when upscaled by the player? I'd be surprised if that is the case. All players I have ever seen resize both 720 and 704 to the same exact 4:3 ratio which is 1440x1080. So the two should not be identical. Although, yeah, the difference is not obvious and small enough for most not to care about. I don't want to get pedantic about it, but while we are at it.


Btw, for analog output it does not matter if the encoded frame size is 704 or 720 as they are both output with the same base clock of 13.5 MHz. What matters in that case is that the scaling is following ITU Rec.601 in the first place, which it most likely does from the nature and age of this TV series.

[jagabo]

Here are some images you can use to make DVD test videos.

720x480:

[Attachment 65449 - Click to enlarge]

704x480:

[Attachment 65450 - Click to enlarge]

The reticles around the edges let you detect if anything is being cropped. When burned as 4:3 DVD the white blocks in the center should be displayed as squares on the digital upscaled outputs. The full frames should be 4:3.

[mistateo]

Wow, mistateo you really did go down the rabbit hole. It took me years to ingest and process the information on the pages you linked to (and do tests on my own to see the actual effects).

In any case, to summarize, my source is what it is, round looks round, and after straight resize from 720x576 to 720x480 round still looks round on my (digital) TV, whether cropping to 704 or keeping the original 720.

Does this mean that both, 720 and cropped to 704, look identical when upscaled by the player? I'd be surprised if that is the case. All players I have ever seen resize both 720 and 704 to the same exact 4:3 ratio which is 1440x1080. So the two should not be identical. Although, yeah, the difference is not obvious and small enough for most not to care about. I don't want to get pedantic about it, but while we are at it.


Btw, for analog output it does not matter if the encoded frame size is 704 or 720 as they are both output with the same base clock of 13.5 MHz. What matters in that case is that the scaling is following ITU Rec.601 in the first place, which it most likely does from the nature and age of this TV series.

There are some inherent problems with my test. As this footage was manually animated in the 80s without computers etc, there are no guarantees on the roundness of a "dragon ball" to begin with even on the original film, pre ntsc (or ntsc->pal -> dvd) conversion. Plus, there are no guarantees I hit the "center" of the circle when measuring millimeters with a wooden ruler vertically and horizontally on my tv to compare, as there were no reference points. I was just saying via my very UNscientific test, it appeared to be the same width/height. To make matters worse, I had left the mode in 480p after testing something else the previous night, so I wasn't testing what I THOUGHT I was testing in the first place. I will retry my very unscientific test with upscaling on, but I don't think it will be valuable as a reference. Edit: On second thought, despite the dvd player being set to 480, the height was still full, so it seems the "fit to screen" feature on the samsung tv will scale the resolution anyway.

I am thinking two things: One, the images posted by jagabo would answer the 720 vs 704 question MUCH more definitively than my very flawed test. Two, someone (probably multiple people, possibly even some in this very thread) likely have already done this test and know the exact outcome, whether upscaled to 1080 or left at 480 over hdmi, or analog signal via composite video, component video, s-video, etc.

But I am totally willing to give this a try and post the result for posterity. What is the recommended way of doing "still picture" dvds? My first thought was just to do them as two different menus. But DVDLab pro 2 will only allow me to use 720x480 for a background image or a 720x540 PSD file, so the 704 frame will be problematic. I could maybe try out the "slide show" feature. Or if brute force requires, I can make mpeg 2 movie out of the same still frame repeated for each size. Any recommendations?

[Sharc]

Um, it does move back to the left after each scan line, and back to the top after scanning the entire field, doesn't it?

Yes. Your pictures apply for PAL but look similar for NTSC.

If you look at the analog video signal with an oscilloscope you will see that during the retrace periods the horizontal and vertical sync information is transmitted (sync pulses etc.). The horizontal line blanking interval includes the so called color burst which is the reference signal for the hue in color TV.

Edit:
See for example here for more details:
https://www.itu.int/dms_pubrec/itu-r/rec/bt/R-REC-BT.470-6-199811-S!!PDF-E.pdf

Edit2:
And the figure below indicates the nominal timing values of a scanline for NTSC composite video. It is in line with the nominal values of the referenced BT.470-6 standard.

Interestingly, the Active Video is shown as 52.6us (actually 53.5556-1.5-4.7-0.6-2.5-1.6 = 52.6556)us which makes 710.85 pixels at 13.5MHz sampling rate according Rec.601. So neither 704 nor 720. Still wondering that some DVDs have active picture widths somewhere between 704 and 720 pixels (padded with tiny bars to fill the 720)?

[mistateo]

Interestingly, the Active Video is shown as 52.6us (actually 53.5556-1.5-4.7-0.6-2.5-1.6 = 52.6556)us which makes 710.85 pixels at 13.5MHz sampling rate according Rec.601. So neither 704 nor 720. Still wondering that some DVDs have active picture widths somewhere between 704 and 720 pixels (padded with tiny bars to fill the 720)?

This is what I have been saying about my source video. Despite it being a PAL DVD, it began as NTSC (in Japan). When I look at the unaltered source in virtualdub, snap the source frame and open in photoshop, it is 720x576, but the left 5 pixels are not active, the right 4 are not active and the last "active" column on the right "doesn't look quite right". The pixels are "active" but they don't make a clean/sharp edge. This coincides exactly with the expected active width for (analog) NTSC.

According to what I have read, there are are 3 different pixel aspect ratios for 4:3 content:

Generic (for native digital content, all 720 pixels active): NTSC 8/9, PAL 16/15.

H.264 (Mpeg 4, 704 pixels active): NTSC 10/11, PAL 12/11.

ITU-R BT.601 (analog origin, 710.85 active pixels NTSC, 702 active pixels PAL): NTSC 38800/42651, PAL 1150/1053

Every one of these standards result in perfect 4:3 ratio when considering the following formula: ActiveWidth * PAR / VerticalResolution. The ITU calculation is the most interesting because vertical resolution is 1/2 + 484 + 1/2 = 485 for NTSC and 1/2 + 574 + 1/2 = 575 for PAL, where as the other two standards just count each half line as whole for pal, and for NTSC 5 lines are just removed when converted to digital.

So it seems like for analog outputs, the dvd player will have to use the ITU spec, and nobody is the wiser due to overscan. Based on conversations above, it seems like when connected via hdmi and upscaled 2.25x on each axis to 1440x1080 the H.264 standard is used. Is there EVER a time where the "Generic" (NTSC 8/9, PAL 16/15) PAR is used by a DVD player? Since it seems like the digital video output of a dvd player (hdmi, but possibly DVI?) is practically always connected to a high definition source, would it just use the H.264 PAR when upscaling, or is it smart enough to use the generic PAR if ALL pixels are active?

[Sharc]

.... Is there EVER a time where the "Generic" (NTSC 8/9, PAL 16/15) PAR is used by a DVD player?

Remember that the DVD standard does not explicitly define a PAR. It only specifies the DAR (DisplayAspectRatio) as 4:3 or 16:9. The PAR is an implicit consequence only which we can calculate depending how we stretch or squeeze the original undistorted picture.
See jagabo's 2 pictures in post #63. The 704x480 picture has an implicit PAR of 10:11, and the 720x480 picture has an implicit PAR of 8:9 as you can measure from the rectangle in the picture. It's nowhere written, we just know it "by design". The DVD player does not (cannot) read this PAR. The player just outputs the 704 x 480 or the 720x480 source as 4:3 when flagged accordingly. It's the responsibility of the author to format (i.e. scaling wxh, adding padding borders as required) the active content such as to look undistorted when played back at the specified DAR (4:3 in your case).
If you have the chance make 2 DVDs from jagabo's 2 pictures and see how the square looks on your TV.

And yes, the Generic PAR applies for 720 wide DVD pictures, typical for 'widescreen' movies based on film sources rather than videotapes or analog sources. They are authored as 720x480 DVDs (720x576 for PAL) and flagged as 16:9. The PAR becomes 32:27 for NTSC and 64:45 for PAL. Letterboxing is used as needed to fit the movie aspect ratio (16:9, 2.35:1, 1.85:1 ...) correctly into the 16:9 playback frame.

For 4:3 the Generic PAR is (implicitly) also used when we pad 704 to 720 for example. The player does not distinguish between the useful 704x480 picture and the pillars. It's just a 720x480 frame for the player which it scales to 4:3. Means the 704x480 useful picture gets ~2% horizontally squeezed.

mpeg4 /AVC H.264 is different from DVD. One can flag the PAR (called SAR Sampling Aspect Ratio in this case to add to the confusion ....) , and the player can play the picture according to the PAR, or one can force to play it at a fix DAR like 4:3 or 16:9. MPC-HC for example has these options.

Edit:
Attached 3 video files, made from jagabo's pictures in post #63 which you should be able to author as DVDs with your usual toolset and see what you get on your TV. Keep an eye on the white rectangle and on the overall picture aspect ratio (w/h ratio)
The file (704+16)x480.mpg is the padded version of the 704x480.mpg. Can you see the slightly squeezed rectangle?

[jagabo]

Interestingly, the Active Video is shown as 52.6us (actually 53.5556-1.5-4.7-0.6-2.5-1.6 = 52.6556)us which makes 710.85 pixels at 13.5MHz sampling rate according Rec.601. So neither 704 nor 720.

Keep in mind that the active picture height is 485 scanlines for NTSC. If you compensate for that 704*485/480 you get very close to 710 pixels. And 704 is really an approximation. It's really 703.something, rounded up to 704 for digital convenience.

[Sharc]

Interestingly, the Active Video is shown as 52.6us (actually 53.5556-1.5-4.7-0.6-2.5-1.6 = 52.6556)us which makes 710.85 pixels at 13.5MHz sampling rate according Rec.601. So neither 704 nor 720.

Keep in mind that the active picture height is 485 scanlines for NTSC. If you compensate for that 704*485/480 you get very close to 710 pixels. And 704 is really an approximation. It's really 703.something, rounded up to 704 for digital convenience.

Yep! Thanks for pointing this out with the 485 scanlines. Everything fits again. We don't need another inconsistency.

[Skiller]

So the way this this series went might be:

• Film
• Telecine to analog NTSC broadcast tape


Some time later:

• Digitization to (most likely) Digital Betacam which stores a 720x486 frame, of which 711x485 maximum would be active picture
• Standards conversion using field blending; 6 scanlines get cut off, resize to 720x576
• PAL DVD of which about 711x575-576 are active (half lines sometimes get blanked)

[Sharc]

Nice conclusion skiller, makes sense to me.

[jagabo]

Some time later:

• Digitization to (most likely) Digital Betacam which stores a 720x486 frame, of which 711x485 maximum would be active picture
• Standards conversion using field blending; 6 scanlines get cut off, resize to 720x576
• PAL DVD of which about 711x575-576 are active (half lines sometimes get blanked)

More likely an analog NTSC tape was run through a standards converter to make an analog PAL tape with field blending. Then the PAL tape was digitized.

[mistateo]

Attached 3 video files, made from jagabo's pictures in post #63 which you should be able to author as dvds with your usual toolset and see what you get on your tv. Keep an eye on the white rectangle and on the overall picture aspect ratio (w/h ratio)
the file (704+16)x480.mpg is the padded version of the 704x480.mpg. Can you see the slightly squeezed rectangle?

thank you!

[Skiller]

You are right, I did not consider that Dragonball came to Europe as early as 1988 by the French (took another 11 years before it arrived in Germany, when Digital Betacam was the standard, so that got me off). Anyways, doesn't really make a difference.


However, if I remember correctly, the French also did their own film transfer which resulted in 25p. At least I am pretty sure the episodes aired on German TV were 25p (no field blending), except for the intro, and they were based on the French version.


But looking at the Dragonball logo in post #19, this seems to be the original Japanese film transfer, thus the field blending.

[mistateo]

I am still at work (uhhhhh, waiting for compiling to finish, haha) so when I get home I will burn the movies to disc to confirm. Now that I know the PAR is non-existent to the DVD (output is 4:3 or 16:9 DAR, regardless of 704 vs 720) this all seems to make more sense. If the resolution is 720x480, 8:9 PAR is implied. If the resolution is 704x480, 10:11 PAR is implied. If that is correct, THIS is what I would expect to see when upscaled to 1440x1080 with simulated overscan OFF:

Native 720x480:
720 x 8/9 * 2.25 = 1440, 480 * 2.25 = 1080, same exact aspect ratio, square is square, pixel at zero markers visible at very edge of the frame

Native 704x480:
704 * 10/11 * 2.25 = 1440, 480 x 2.25 = 1080, same exact aspect ratio, square is square, pixel at zero markers visible at every edge of the frame

704x480 padded to 720x480:
720 * 8/9 * 2.25 = 1440, 480 x 2.25 = 1080, BUT the ACTIVE area being only 704x480...
Which means 704 * 8/9 * 2.25 = 1408, 480 x 2.25 = 1080. So this final image size: 1408x1080 is obviously different than the other two of 1440x1080. Since this isn't a proper resolution, I expect the black bars on the sides to be 16px wider on each side. So in this case I expect the square to be taller than it is wide, with the 0 pixel markers still visible, but the distance between the left and right 0 markers will be LESS than the other two movies.

My TV is 46" 16x9. If I am to trust that this size yields 47.889 pixels per inch (according to this site): then the distance between the zero markers for the non-padded videos should be (1440/47.889) about 30.07 inches apart. The distance between zero markers for the padded video should be (1408/47.889) about 29.40 inches apart, while the distance between top and bottom zero markers should remain consistent in all 3 videos. Thus, a second measure of proof that 704 padded to 720 will indeed distort the original aspect ratio.

If my hypothesis above is wrong, or not even remotely right, I will request the men in black come to erase my memory to the point before I started investigating all of this resolution madness...

[Sharc]

704x480 padded to 720x480:
720 * 8/9 * 2.25 = 1440, 480 x 2.25 = 1080, BUT the ACTIVE area being only 704x480...
Which means 704 * 8/9 * 2.25 = 1408, 480 x 2.25 = 1080. So this final image size: 1408x1080 is obviously different than the other two of 1440x1080.

Yes. The 1408/1440=0.9778 is the often cited ~2% error.

Here the padded file again. It is the same as before but the pads are green now which will make them visible in the dark background of the TV screen.