DVB 544x576 mode - pixel aspect ratio and centre point?

The DVB standards (in particular, ETSI TR 101 154) define that
544x576 is a valid MPEG-2 luminance sampling grid resolution for
DVB broadcasts.

The standards also suggest that if 544x576 is to be converted
(upsampled) to ITU-R BT.601 pixels, the figure 544 needs to be
multiplied by 4/3.

However, this calculation will give 725.333... ITU-R BT.601
pixels, which is a tad more than the typical framebuffer
width of 720 or 704 pixels. Using all those pixels in real
world equipment would mean extending into the horizontal
blanking period, not to mention using a wider framebuffer
than is usually used! Obviously, something needs to give in.

Now, is the intention of the 544x576 format that only a
540 pixel wide area of the sampling grid is actually used,
and maps to 720 ITU-R BT.601 pixels (and the extra 4 pixels
are there just to satisfy the MPEG-2 needs-to-be-evenly-
divisible-by-16 technicalities, and can be cropped off
before the conversion?) This would make sense, since
540 * 4/3 = exactly 720 (and 540 is not evenly divisible
by 8 or 16, but 544 is.)

If the above assumption is correct, where is the centre point
located in this 544x576 image?

For example, if I want to convert a 720x576 (13.5 MHz) image
to the 544x576 format, do I first downsample to 540x576 and
then add 2 pixel columns to the left and the right side (making
the middle point in the original picture the middle point of
the 544x576 picture), or do I just add a 4 pixel column on the
right side (in which case the idea would be that the rightmost
4-pixel column is just slack space for the MPEG-2 divisible-by-16
requirements)?

Is there any established, standard way of doing it?

(I am asking this because it has been suggested that I should
add this common DVB resolution to my aspect ratio guide &
conversion table at http://www.iki.fi/znark/video/conversion/,
but I am not exactly sure if my above-mentioned assumptions
about this sampling grid format are valid.)

--
znark

"Jukka Aho" wrote in message
...
The DVB standards (in particular, ETSI TR 101 154) define that
544x576 is a valid MPEG-2 luminance sampling grid resolution for
DVB broadcasts.

The standards also suggest that if 544x576 is to be converted
(upsampled) to ITU-R BT.601 pixels, the figure 544 needs to be
multiplied by 4/3.


Am I right in thinking this is the resolution for a 4:3 picture broadcast
12P16?

Mat Overton wrote:

544x576

Am I right in thinking this is the resolution for a
4:3 picture broadcast 12P16?

544x576 is supposed to be stretched over the full width of
16:9 or 4:3 screen, so this is more like a bandwidth-saving,
slightly lower resolution format for the penny-pinching
broadcasters.

--
znark

Mat Overton wrote:
"Jukka Aho" wrote in message
...
The DVB standards (in particular, ETSI TR 101 154) define that
544x576 is a valid MPEG-2 luminance sampling grid resolution for
DVB broadcasts.

The standards also suggest that if 544x576 is to be converted
(upsampled) to ITU-R BT.601 pixels, the figure 544 needs to be
multiplied by 4/3.


Am I right in thinking this is the resolution for a 4:3 picture
broadcast 12P16?

I believe this is close to being the case.

However this resolution is ALSO used as a full-width DVB standard, indeed
quite a few DTT and DSat services use it for both 4:3 and 16:9 services.
(AIUI all ITV1 variants on DSat are 544x576 resolution - and they broadcast
full frame 4:3 and full frame 16:9 on this platform) On DTT I think ITV1 is
720/704x576 (can't recall which) but the ITV News Channel (4:3 only) is
544x576, as is bid.up TV I think (16:9 only)

Steve

Jukka Aho wrote:
The DVB standards (in particular, ETSI TR 101 154) define that
544x576 is a valid MPEG-2 luminance sampling grid resolution for
DVB broadcasts.

The standards also suggest that if 544x576 is to be converted
(upsampled) to ITU-R BT.601 pixels, the figure 544 needs to be
multiplied by 4/3.

However, this calculation will give 725.333... ITU-R BT.601
pixels, which is a tad more than the typical framebuffer
width of 720 or 704 pixels. Using all those pixels in real
world equipment would mean extending into the horizontal
blanking period, not to mention using a wider framebuffer
than is usually used! Obviously, something needs to give in.

Now, is the intention of the 544x576 format that only a
540 pixel wide area of the sampling grid is actually used,
and maps to 720 ITU-R BT.601 pixels (and the extra 4 pixels
are there just to satisfy the MPEG-2 needs-to-be-evenly-
divisible-by-16 technicalities, and can be cropped off
before the conversion?) This would make sense, since
540 * 4/3 = exactly 720 (and 540 is not evenly divisible
by 8 or 16, but 544 is.)

If the above assumption is correct, where is the centre point
located in this 544x576 image?

For example, if I want to convert a 720x576 (13.5 MHz) image
to the 544x576 format, do I first downsample to 540x576 and
then add 2 pixel columns to the left and the right side (making
the middle point in the original picture the middle point of
the 544x576 picture), or do I just add a 4 pixel column on the
right side (in which case the idea would be that the rightmost
4-pixel column is just slack space for the MPEG-2 divisible-by-16
requirements)?

Is there any established, standard way of doing it?

(I am asking this because it has been suggested that I should
add this common DVB resolution to my aspect ratio guide &
conversion table at http://www.iki.fi/znark/video/conversion/,
but I am not exactly sure if my above-mentioned assumptions
about this sampling grid format are valid.)

My gut instinct would be to centre the 540 within the 544, in the same way
that the 702 samples are centred within the 720. (702 samples being the
4:3/16:9 active area - with the extra 18 adding a little extra width in
digital systems) I believe that the 702 samples are centred within the 704
sample variant as well?

Steve

Stephen Neal wrote:

My gut instinct would be to centre the 540 within the 544, in
the same way that the 702 samples are centred within the 720.

Yes, this would sound logical.

The digital active image area of 544 pixels @ 10.125 MHz being
even wider than 720 pixels @ 13.5 MHz just made me suspect that
maybe there is some special way of handling these images. (Also,
I have seen a couple of 544x576 DVB streams where the image had
a wider right border than the left one, but maybe this was just
coincidental.)

I believe that the 702 samples are centred within the 704
sample variant as well?

Can't say I would have seen any official guide as for how to
handle this, but as there is an EBU recommendation about
centering 702x576 in the middle of 720x576, I have always
believed that the same principle applies to 704 pixel wide
13.5 MHz sampling grids, too.

Who writes these standards, anyway, and why don't they document
these things better? What do they get paid for?

--
znark

"Jukka Aho" wrote in message
...
Can't say I would have seen any official guide as for how to
handle this, but as there is an EBU recommendation about
centering 702x576 in the middle of 720x576, I have always
believed that the same principle applies to 704 pixel wide
13.5 MHz sampling grids, too.

Who writes these standards, anyway, and why don't they document
these things better? What do they get paid for?

The real question is what do you have to smoke to devise standards whether
the ratio of pixels in the horizontal and vertical directions is neither 4:3
nor 16:9? Does it not make eminent sense to have the same resolution
(pixels/mm) in both directions at least for *one* of 4:3 and 16:9?

But then the obscure numbers that have become enshrined in standards have
always intrigued me. Why are audio tape speeds not integer numbers of inches
per second (eg 2, 4, 8, 16 etc) or else round numbers of cm/sec? Why is the
UK railway gauge 4' 8½" rather than 4' 6", 4' 9" or 5' (or else round
numbers of centimetres)? Why are computer drives either 5¼" or 3½" rather
then 5" and 3"? Why is there not an integer number of cubic inches in a
gallon (so as to relate linear and volumetric measurement)? Why are all
imperial units related by factors other than ten (eg inches/foot,
yards/mile, yards/chain, pounds/stone etc)? Why is cinema film shot at 24
frames/sec rather than 20, 25 or 30, given that 50Hz (halved to derive 25
Hz) and 60Hz (halved to derive 30 Hz) are the standard mains frequencies?

The weirdest one is the spec for high-definition TV which uses 1080 lines -
neither 2xPAL (1250) nor 2xNTSC (1050) lines, so *both* standards will have
to be interpolated using weird conversion factors (leading to loss of
sharpness) when showing old low-definition material on high-definition.

Maybe I just like numerical simplicity: if I had a clean sheet of paper to
devise a standard, I'd always ensure that the numbers that it used were nice
and simple in whatever system of units (presumably SI) that I used.

Going further off-topic, why are many rev counters in cars labelled with
numbers like 10, 20, 30 x 100 rpm rather than the engineering standard of
expressing small or large numbers as powers of 1000 (eg 1, 2, 3 x 1000
rpm) - so as to express them as numbers in the range 1-999.999 x 1/1000000,
1/1000, 1, 1000, 1000000 etc?

On Sat, 28 Feb 2004 18:50:00 GMT, "Martin Underwood"
wrote:

| Why is the
| UK railway gauge 4' 8½" rather than 4' 6", 4' 9" or 5' (or else round
| numbers of centimetres)?

That one at least is easy, it depends on the width of the bums of two
horses. The original ?railway? trucks were ordinary carts, with wheels
4" 8 1/2" pulled by horses, along wooden rails. Wooden rails were much
smoother than cart tracks. As improvements like iron rails, and steam
locomotives the gauge stayed the same. Roman roads have ruts of much the
same gauge.

http://www.spartacus.schoolnet.co.uk/RAgauge.htm

Isombard Kingdom Brunell had a great idea when he designed the Great
Western Railway. He built it in the technically much better Broad Gauge
2.2m but the standard won out against the better.

You can see a replica of the broad gauge Iron Duke locomotive at National
Railway Museum in York.

All standards happened in a similar sort of way :-(((((((((

--
Dave Fawthrop dave hyphenologist co uk Killfile and Anti Troll FAQs
at http://www.hyphenologist.co.uk/killfile.

Martin Underwood wrote:

The real question is what do you have to smoke to devise
standards whether the ratio of pixels in the horizontal
and vertical directions is neither 4:3 nor 16:9? Does it
not make eminent sense to have the same resolution
(pixels/mm) in both directions at least for *one* of
4:3 and 16:9?

Well, I am afraid everything has its reasons... it is just that
the reasoning behind the choices is not necessarily written down
in the standards (which does not help in understanding them),
and implementation hints are scarce!

Here is quite an interesting article about those ITU-R BT.601/656
numbers:

http://groups.google.com/groups?selm...Gy%40hpqmoea.s
qf.hp.com

More of the same...

http://groups.google.com/groups?selm...%2475a048c0%24
712b7c0a%40pc-l301385.wn.bbc.co.uk

.... and yet more!

http://groups.google.com/groups?selm...%24ef84a7e0%24
652b7c0a%40pc-234866.wn.bbc.co.uk

(The URLs above have been split on two lines. If your newsreader
cannot handle them automatically, please copy them into your
web browser in two parts.)

--
znark

"Martin Underwood" wrote

But then the obscure numbers that have become enshrined in standards have
always intrigued me. Why are audio tape speeds not integer numbers of
inches
per second (eg 2, 4, 8, 16 etc) or else round numbers of cm/sec?

Early wire recorders ran at 60ips (and were VERY dangerous for the operators
if the wire broke, which happened all too frequently). Speeds of tape
followed, and halved as technology allowed, only going "wrong" below 15ips.

The weirdest one is the spec for high-definition TV which uses 1080
lines -
neither 2xPAL (1250) nor 2xNTSC (1050) lines, so *both* standards will
have
to be interpolated using weird conversion factors (leading to loss of
sharpness) when showing old low-definition material on high-definition.

FWIW, the comparison with 1080 should be 1152 for PAL (2x576 active lines)
and 960 for NTSC (2x480), so from NTSC 1080 = 9/4 * 480, and from PAL 1080 =
15/8 * 576. Not such weird conversion factors, so maybe the standards guys
did know something.

John Howells