Aspect Ratio

[David Crossman]

Why is it that PAL video digitises to 720 x 576 pixels?
- this is NOT 4:3.

A TV 16:9 image is also 720x576 because its anamorphic.

I've calculated that the width should be 1024 pixels to
make a 16:9 aspect ratio but I doubt this is correct
due to the anomoly above.

The answer might be in different shaped pixels but I
just can't seem to get my head round the maths.

TIA.

--
David Crossman DGGB
-------------------
http://www.dareks.dircon.co.uk/

[Phil Ronan]

On 30/8/01 9:07 am, David Crossman at dar...@NOJUNK.dircon.co.uk wrote:

> Why is it that PAL video digitises to 720 x 576 pixels?
> - this is NOT 4:3.
>

Digital video systems often use a screen size of 720 x 576. But the pixels
aren't square (15:16)

Phil

--
phi...@mactrombone.com
^^^^^^^^
Remove the musical instrument to reply by email

[Roger Mellie]

"David Crossman" <dar...@NOJUNK.dircon.co.uk> wrote in message
news:ant004...@dircon.co.uk...

> Why is it that PAL video digitises to 720 x 576 pixels?
> - this is NOT 4:3.

You're assuming the pixels are square - but they're not. Also, you're assuming
that horizontal resolution is the same thing as vertical resolution, which it
isn't.

[Stephen Neal]

David Crossman wrote in message ...

>Why is it that PAL video digitises to 720 x 576 pixels?
>- this is NOT 4:3.

The figure for a compatible 4:3 625 line image is actually 702x576 samples.
The 720x576 figure commonly quoted is for the slightly wider (to allow for
timing errors?) digital format. However neither of these are really "PAL" -
though annoyingly PAL seems to be used as a description of line format
nowadays even though such sampling does not any real link with the PAL
subcarrier system.

(And also PAL-M exists, and that is 525 based!!)

>
>A TV 16:9 image is also 720x576 because its anamorphic.

Yes - a choice was made by broadcasters to retain the existing production
infrastructure, based around the existing 13.5MHz luminance sampling, for
16:9 production. This means that the aspect ratio of picture samples is
different between 16:9 and 4:3 formats. Interestingly the original digital
standards also proposed an 18MHz variant for 16:9, which would have
increased the horizontal resolution to 1080 samples per line, and the pixels
in the 4:3 and 16:9 formats would have retained the same "shape" between
formats.

Instead we have not increased resolution, so 4:3 samples are a different
"shape"/aspect ratio to the wider 16:9 samples.

>
>I've calculated that the width should be 1024 pixels to
>make a 16:9 aspect ratio but I doubt this is correct
>due to the anomoly above.
>
>The answer might be in different shaped pixels but I
>just can't seem to get my head round the maths.

Yep - if you keep the samples/line and number of lines the same, but change
the aspect ratio of the source picture, you change the aspect ratio of the
sample area.

[David Crossman]

In an article, Stephen Neal

<stephe...@nospam.as-directed.com> wrote:

> annoyingly PAL seems to be used as a description of line
> format nowadays even though such sampling does not any
> real link with the PAL subcarrier system.

Thanks for your detailed reply. It is just easier to hit
3 keys (PAL) than type 625 lines/25 fps I guess.

> if you keep the samples/line and number of lines the same,
> but change the aspect ratio of the source picture, you change
> the aspect ratio of the sample area.

Yes I see that but in practice: If I have have a properly
proportioned 1024 x 576 image, re-sample the width to 720
and then show this (anamorphic) image on a 16:9 display,
will the image be properly proportioned again?

Or should I be applying Phil Ronan's 15:16 somewhere along
the line(!)?

--
David Crossman DGGB, LRPS
-------------------------
http://www.dareks.dircon.co.uk/

[Andy Woodhouse]

David Crossman <dar...@NOJUNK.dircon.co.uk> wrote in article
<ant004...@dircon.co.uk>...

I have read some of the follow up threads to this posting,
but this seemed the most appropriate "starter" at which to
jump into the discussion. I also aplogise in advance for
a rather lengthy arnd arithmetic response.

Let me first make clear that the term "pixel" or "pel" can
be misleading when dealing with sampled systems. Many text
books which show pixels as little rectangles - which often
causes the reader to assume that the pixel "occupies" the
space covered by that rectangle.

When sampling a signal the aim is to measure the
instantaneous value of the signal property (usually the
voltage) at a defined point in time. The width of the
sampling "pulse" is ideally 0.0 seconds. "Real" sampling
processes in ADC chips try to get as close as possible to
the ideal width. Thus the rectangular grid should really
be thought of a series of sample instants which occur at
(say) the top left corner of the grid rectangles.

The 720 by 576 sample numbering originates in the ITU-R
BT.601 (a.k.a. Rec.601) sampling standard. When this was
researched and designed in the late 1970's there were
(essentially) only two scanning standards, 625/50/2:1 and
525/59.94/2:1. The active line duration of both scanning
standards is very similar, with 625/50 using 52 microseconds
H.A.D. (Half Amplitude Duration), and 525/59.94 at 52.66
microseconds H.A.D. There is some permitted variation on
these times, as the active line length is set by the total
line minus the blanking duration.

ITU-R BT.470 notes that the balnking tolerance in 625/50 PAL
can be plus or minus 300 nanoseconds. (Less than this in
system I as used in the U.K.) Measuring from the start of
rise out of blanking through to the fall back to blanking
extends the above figures by about 0.5 microseconds.

It was clearly sensible, to keep down equipment costs, if a
common sampling frequency could be used for both standards,
and that the number of samples per active line was set to
the same number whichever scanning standard applied.

To make subsequent processing simple and straightforward
(e.g. vision mixing, DVE effects), the equivalent samples
on all lines must occur at the same instant or position
relative to reference sync timing. This sampling structure
occurs when the sample clock is an integer multiple of the
line frequency. The first frequency which is sufficiently
high to meet the practical Nyquist criterion (2.2 x bandwidth)
and which is an integer multiple of 656 and 525 line frequencies
is 13.5 MHz. Hence when people think of "PAL" sampled data they
assume that 702 samples are required. However the finite rise
and fall of blanking requires extra samples be taken if image
trunkation is to be avoided. For 525 line signals this implies
a need to digitise at least 53.16 microseconds of video
(approx 718 samples). Allowing for some permitted tolerance on
line duration the samples per line then rounds up to 720.

If the horizontal and vertical spacing of the sample points was
identical ("square pixels") then we would need (576 * 4 / 3) =
768 samples for a 52 microsecond zero rise time image. Using
13.5 MHz sampling gives 702 samples for the same duration, thus
giving a horizontal sample separation which is 768/702 = 1.094
times the separation of the vertical sampling.

Similarly the number of samples for a "square pixel" 16 by 9
aspect ratio image would require (576 * 16 / 9) = 1024. An ITU-R
BT.601 sampled 16 by 9 image can still use 702 samples per 52
microseconds, thereby giving a horizontal to vertical sampling
distance ratio of 1024/702 = 1.459 (to 3 places!!).

Most computer painting systems use square pixels and therefore to
use such a package to make a still for export to Rec. 601 would
suggest that the "square pixel" image should start at 1024 by 576,
being resampled to 702 by 576, then placed centrally within a 720
by 576 black background. However the instant rise time transition
of blanking could cause severe ringing when an analogue output
signal is created. The better width to use for the "square pixel"
source is (1024 * 720 /702) = 1050 samples. After resampling to
720 by 576 the image can be output from an ITU-R BT.601 system.
Any subsequent DAC process is responsible for applying standard
"analogue system" blanking to the signal, giving a 52 microsecond
H.A.D. line. A 4 by 3 image should start life as a 787 or 788
sample per line "square pixel" source (ideally 787.69 samples!!).

As Stephen Neal noted, there is an option within ITU-R BT.601 for
higher sample clock rates in 16 by 9 picture operations (Part B
to annexe 1). This allows an 18 MHz sample clock for the luminance
signal with 1152 samples per total line and 960 samples per
active line. The figures are produced by a simple scaling of the
numbers used for 4 by 3 in 13.5 MHz sampled operations. Thus samples
per active line becomes ((720 * 16/9) / (4/3)) = (720 * 4 / 3).

As many threads in this newsgroup have demonstrated there are
enough problems with 4 by 3 and 16 by 9 interworking. Adding the
new sample frequency would have significantly increased the
complexity and hence cost of the production and post-production
systems. The suggestion of high sample frequency operation emerged
at about the same time as accountants became the driving force in
broadcasting. I know that I get more cynical as I get older, but
it seems plausible that staying with 13.5 MHz may have had
something to do with the money men. However even if we had chosen
18 MHz sampling we would probably have returned to 13.5 MHz at the
input to the MPEG compression equipment as Main Profile Main Level
used in DVB supports a maximum of 720 samples per line. Using a
higher level would significantly increase the cost of the decoder
box, with little or no improvement in perceived picture quality,
even if there were no pressure to use minimum data rates.

Given a vertical sampling system which uses 576 sample points the
maximum rate of change signal would have 288 cycles in the height
of the picture. However the use of interlace reduces the usable
resolution, but we know that we can use more than 144 cycles per
picture height (limit defined by 288 samples per field), without
flicker becoming too annoying. Some work was done to establish what
effective vertical resolution was available on modern displays.
Once that figure was established, and assuming that the horizontal
and vertical resolutions should be almost identical lead to a
horizontal bandwidth requirement of 5.7 MHz, which is just within
the bandwidth available when using 13.5 MHz sampling.

Andy Woodhouse
Lecturer
BBC Training & Development.

=================================================================
Note : All views expressed are my own, and may differ from those
of my employer.

[jim]

>
>Yes I see that but in practice: If I have have a properly
>proportioned 1024 x 576 image, re-sample the width to 720
>and then show this (anamorphic) image on a 16:9 display,
>will the image be properly proportioned again?

The BBC says no to this - the rest of the 625 line world says yes.

The BBC point out that the non-squareness is derived from the 702
pixels sitting within the 720 pixel image.

(16 ÷ 9) x 576lines x (720 ÷ 702) = 1050 square pixels wide

jim

[ne...@rtrussell.co.uk]

jim <jim.howatDE...@ntlworld.com> wrote:

: The BBC says no to this - the rest of the 625 line world says yes.

: The BBC point out that the non-squareness is derived from the 702
: pixels sitting within the 720 pixel image.

: (16 / 9) x 576lines x (720 / 702) = 1050 square pixels wide

It's by no means true that the BBC is at odds with the
rest of the 625-line world. That the 4:3 (or 16:9)
image consists only of the central 702 pixels is quite
clearly specified in EBU Technical Recommendation R92-
1998: Active line length in analogue and digital 625/50
television systems:

"The EBU recommends that in 625-line television systems
sampled to ITU-R Rec. BT.601 part A, only the central
702 samples of the digital active line (samples 9-710
inclusive) are used to carry the active picture. The
remaining 18 samples should not be used to carry any
picture information that is intended for display
either for 4:3 or for 16:9 aspect ratio images".

You can find this at http://www.ebu.ch/pmc_r92-1998.pdf

Richard.
http://www.rtrussell.co.uk/

[jim]

Richard,
My "at odds" reply was an answer to the square pixel equivalent
question.

careful what you snip when quoting... :)

cheers

jim

[tommym...@gmail.com]

[Roger Wilmut]

In article <216f3c2e-f2d2-48ea...@googlegroups.com>,

4 x 3 is very nominal. 625 line transmissions use only 576 lines for
actual picture content (leaving some blank lines for the flyback and
some in use for Teletext and internal switching information); digital SD
transmissions match vertical pixels to lines. The aspect ratio of 625
line is slightly wider than 4 x 3 (plus the fact that TVs routinely
overscanned, i.e. some picture disappeared behind the mask). By the tie
625 lines was introduced cinema films were no longer being made in true
4 x 3, but slightly wider, often 1.66:1.

Widecreen TV on SD digital is a fudge. The pixels are (roughly) 16 x 9
instead of being (roughly) square, and the picture is stretched
laterally to fill the screen, with the result that the horizontal
definition no longer matches the vertical definition. A broadcast flag
tells the set whether to fill the screen or display in 4 x 4 with side
panels (though many consumers prefer to fill the screen, thus making
everyone look fat).

Only with HD does true widescreen occur: the pixels are square so that
the definition is the same in both planes. There is no 4 x 3 mode; films
in that ratio are transmitted with black panels on each side which are
actually part of the transmitted image.

[NY]

"Roger Wilmut" <rfwi...@nospam.yahoo.com> wrote in message
news:rfwilmut-E3C4F9...@news.virginmedia.com...

> 4 x 3 is very nominal. 625 line transmissions use only 576 lines for
> actual picture content (leaving some blank lines for the flyback and
> some in use for Teletext and internal switching information); digital SD
> transmissions match vertical pixels to lines. The aspect ratio of 625
> line is slightly wider than 4 x 3 (plus the fact that TVs routinely
> overscanned, i.e. some picture disappeared behind the mask). By the tie
> 625 lines was introduced cinema films were no longer being made in true
> 4 x 3, but slightly wider, often 1.66:1.

Yes although the resolution of SD is slightly narrower than 4:3 - with 576
pixels/lines vertically you'd expect 768 pixels horizontally whereas there
are only 720 pixels. So SD has to be stretched very slightly widthways to
achieve a 4:3 picture.

> Widecreen TV on SD digital is a fudge. The pixels are (roughly) 16 x 9
> instead of being (roughly) square, and the picture is stretched
> laterally to fill the screen, with the result that the horizontal
> definition no longer matches the vertical definition. A broadcast flag
> tells the set whether to fill the screen or display in 4 x 4 with side
> panels (though many consumers prefer to fill the screen, thus making
> everyone look fat).
>
> Only with HD does true widescreen occur: the pixels are square so that
> the definition is the same in both planes. There is no 4 x 3 mode; films
> in that ratio are transmitted with black panels on each side which are
> actually part of the transmitted image.

Actually the black panels either side of a 4:3 picture is done on SD too,
though different broadcasters have different policies: BBC almost always
transmits 4:3 cinema films and often transmits 4:3 made-for-TV programmes in
a 16:9 frame with black side borders, with the widescreen flag set to ON, so
the resolution of the picture itself is a lot less than 720 pixels, whereas
most other channels usually transmit 4:3 material so as to fill the whole
720 width and set the widescreen flag accordingly to OFF.

[Brian Gaff]

Are there not wasted lines in most Pal transmissions as well?

Brian

--
From the Sofa of Brian Gaff Reply address is active
<tommym...@gmail.com> wrote in message
news:216f3c2e-f2d2-48ea...@googlegroups.com...
Le jeudi 30 ao�t 2001 10:25:42 UTC+2, David Crossman a �crit :

[Stephen]

"Brian Gaff" <Bri...@blueyonder.co.uk> wrote in message
news:l45t3g$2fu$1...@dont-email.me...

> Are there not wasted lines in most Pal transmissions as well?
>
> Brian

Yes. 50 blank lines per frame in analogue PAL (625 minus 575), and 49 in
Standard Definition digital (625 minus 576). It's 45 in HD (1125 minus 1080)
as some HD formats transmit the blanked lines. 1080i HD is actually a TV
system with 1125 lines.

> --
> From the Sofa of Brian Gaff Reply address is active
> <tommym...@gmail.com> wrote in message
> news:216f3c2e-f2d2-48ea...@googlegroups.com...
> Le jeudi 30 ao�t 2001 10:25:42 UTC+2, David Crossman a �crit :
>> Why is it that PAL video digitises to 720 x 576 pixels?
>> - this is NOT 4:3.
>>
>> A TV 16:9 image is also 720x576 because its anamorphic.
>>
>> I've calculated that the width should be 1024 pixels to
>> make a 16:9 aspect ratio but I doubt this is correct
>> due to the anomoly above.
>>
>> The answer might be in different shaped pixels but I
>> just can't seem to get my head round the maths.
>>

The reason for 720 instead of 768 is because they chose to base the SD
digital standard on a 13.5 MHz sample clock, and that's how the numbers work
out. 720 clock pulses in each active line period (the period with active
picture information in it).

[Roger Wilmut]

In article <YJCdnZ5GCMUN7_vP...@brightview.co.uk>,

"NY" <m...@privacy.net> wrote:

> Actually the black panels either side of a 4:3 picture is done on SD too,
> though different broadcasters have different policies: BBC almost always
> transmits 4:3 cinema films and often transmits 4:3 made-for-TV programmes in
> a 16:9 frame with black side borders, with the widescreen flag set to ON, so
> the resolution of the picture itself is a lot less than 720 pixels, whereas
> most other channels usually transmit 4:3 material so as to fill the whole
> 720 width and set the widescreen flag accordingly to OFF.

I've seen that done but only very occasionally, and they shouldn't be
doing it - pure laziness. However mostly when I've been watching old
films on SD the proper ratio is used.

[Mark Carver]

Roger Wilmut wrote:
> In article <YJCdnZ5GCMUN7_vP...@brightview.co.uk>,
> "NY" <m...@privacy.net> wrote:
>
>> Actually the black panels either side of a 4:3 picture is done on SD too,
>> though different broadcasters have different policies: BBC almost always
>> transmits 4:3 cinema films and often transmits 4:3 made-for-TV programmes in
>> a 16:9 frame with black side borders, with the widescreen flag set to ON, so
>> the resolution of the picture itself is a lot less than 720 pixels, whereas
>> most other channels usually transmit 4:3 material so as to fill the whole
>> 720 width and set the widescreen flag accordingly to OFF.
>
> I've seen that done but only very occasionally, and they shouldn't be
> doing it - pure laziness.

4:3 programmes whizz around the Beeb and RedBee playout internally,
pillarboxed (12P16) anyway, so that's the point the horizontal resolution gets
stuffed.

The Beeb transmit that pillarboxed signal on DTT, but then flag the receiver
to output (STBs) or display (TVs) 4:3 full screen by use of AFDs

On D-Sat AFDs are not used (because BSkyB don't support them) so 4:3 pillarbox
programmes are ARC'd back into full frame 4:3 (12F12), effectively rescaling
the 500ish active pixels back to 720, so that's a double whammy !

All UK broadcsters do much the same.

--
Mark
Please replace invalid and invalid with gmx and net to reply.

[NY]

"Roger Wilmut" <rfwi...@nospam.yahoo.com> wrote in message

news:rfwilmut-AD935D...@news.virginmedia.com...

Most black and white films on BBC2 or BBC Four (eg Whistle Down the Wind,
The Wooden Horse) are broadcast as a 4:3 picture embedded in a 16:9 frame.

Likewise when old episodes of classic 1970s programmes are shown (Dad's Army
etc) they are often (though not always) shown this way.

ITV, CH4, Yesterday and Drama get it right: they religiously toggle the
widescreen flag between programme (4:3) and adverts (16:9) at every advert
break, which is how it should be done.

[Paul Ratcliffe]

On Tue, 22 Oct 2013 12:18:07 +0100, Roger Wilmut <rfwi...@nospam.yahoo.com>
wrote:

> In article <216f3c2e-f2d2-48ea...@googlegroups.com>,
> tommym...@gmail.com wrote:
>
>> Le jeudi 30 août 2001 10:25:42 UTC+2, David Crossman a écrit :
>> > Why is it that PAL video digitises to 720 x 576 pixels?
>> > - this is NOT 4:3.

YES IT IS. The pixels ain't square.

> The aspect ratio of 625 line is slightly wider than 4 x 3

No it isn't. 51.95us and 574 + 2 half lines is (or was) analogue 4x3.

> Widecreen TV on SD digital is a fudge.

No more of a fudge than anything previous. The Kell factor was always the
justificatin for this.

> The pixels are (roughly) 16 x 9 instead of being (roughly) square

No they're not.

> and the picture is stretched laterally to fill the screen

No it isn't.

> with the result that the horizontal
> definition no longer matches the vertical definition.

It never did. See above.

[Paul Ratcliffe]

On Thu, 24 Oct 2013 17:50:19 +0100, Mark Carver <mark....@invalid.invalid>
wrote:

> 4:3 programmes whizz around the Beeb and RedBee playout internally,
> pillarboxed (12P16) anyway, so that's the point the horizontal resolution gets
> stuffed.
>
> The Beeb transmit that pillarboxed signal on DTT, but then flag the receiver
> to output (STBs) or display (TVs) 4:3 full screen by use of AFDs
>
> On D-Sat AFDs are not used (because BSkyB don't support them) so 4:3 pillarbox
> programmes are ARC'd back into full frame 4:3 (12F12), effectively rescaling
> the 500ish active pixels back to 720, so that's a double whammy !

They were f'ing about with AFDs during Floyd on 2 this afternoon. Saw it
switch at least 3 times. AYBS was OK afterwards, so I dunno what the problem
was.

[Paul Ratcliffe]

On Tue, 22 Oct 2013 13:42:48 +0100, NY <m...@privacy.net> wrote:

> Yes although the resolution of SD is slightly narrower than 4:3 - with 576
> pixels/lines vertically you'd expect 768 pixels horizontally whereas there
> are only 720 pixels. So SD has to be stretched very slightly widthways to
> achieve a 4:3 picture.

You are just dividing numbers. You don't really understand what you are
talking about.
720x576 can contain both 4x3 full screen and 16x9 full screen pictures.
Bet that blows your mind...

[NY]

"Paul Ratcliffe" <ab...@orac12.clara34.co56.uk78> wrote in message
news:slrnl6j4pn...@news.pr.network...

No it doesn't.

It would make sense for one of the two aspect ratios to use square pixels:
ie to use 768 rather than 720 pixels widthways so the pixels/mm resolutions
are the same on both axes for SD pictures.

But as I now realise, thanks to an earlier posting, the overriding factor is
that they wanted to use a 13.5 MHz pixel clock, which results in 720 pixels
and a slightly lower resolution on the horizontal rather than vertical
axis - a very slightly stretched picture to match a 720x576 pixel image to a
screen than has the same pixel spacing in both axes (eg a 768x576 screen -
always assuming a 1:1 correspondence between picture and screen pixels,
which I'm sure there isn't!).

For 16:9, everything is stretched: the horizontal resolution in pixels/mm is
much lower - a square-pixel system would have 1024x576 pixels. But that's
fair enough.

I suppose in an ideal world you'd design TV screens so they have a physical
pixel spacing on the two axes which matches the most widely-used TV system
that it will be required to display, to avoid interpolation artefacts. But
how do you define "widely-used TV system" when there's 544x576
(reduced-bandwidth channels like Yesterday), 704x576 (SD on ITV), 720x576
(SD on most other channels), 1280x720 (720p HD), 1440x1080
(reduced-bandwidth 1080p HD) and 1920x1080 (full 1080p HD) - and probably
others. And I'm concentrating on "PAL" standard rather than "NTSC" standard;
I wonder if TVs for the two markets are fitted with the same screen.

I suppose TV manufacturers could fit a screen which has a native resolution
which matches the highest broadcast format that will be encountered
(1920x1080) and let it interpolate for all lower resolutions - which is fine
until 4K television becomes the standard a few decades from now!

[Andy Furniss]

NY wrote:

> But as I now realise, thanks to an earlier posting, the overriding
> factor is that they wanted to use a 13.5 MHz pixel clock, which results
> in 720 pixels

Just to add to the confusion, I was told (on here) that for SD it's
actually the central 704 pixels that represent the 4/3 or 16/9 image -
the edge pixels in 720 as broadcast are often but not always blank.

[Roger Wilmut]

In article <BPadnfhOLME0FPTP...@brightview.co.uk>,

"NY" <m...@privacy.net> wrote:

> Most black and white films on BBC2 or BBC Four (eg Whistle Down the Wind,
> The Wooden Horse) are broadcast as a 4:3 picture embedded in a 16:9 frame.

It seems pretty random. Both the RKO B-movies early on Saturday mornings
and the Cagney and Lacey reruns are being shown properly in 4x3.

[Paul Ratcliffe]

On Fri, 25 Oct 2013 11:43:07 +0100, Andy Furniss <an...@andromeda64.plus.com>

702 not 704 actually. I posted a first principles derivation of the 13.5MHz
clock frequency some months back...

http://newsgroups.derkeiler.com/Archive/Uk/uk.tech.broadcast/2013-03/msg00150.html

[Andy Furniss]

Ahh, yes, that's the thread that I got 704 from.

Having just looked at a recording from BBC2 last night, it does seem
that the BBC do, as mentioned in the thread, round up to 704 to be mpeg
friendly - the black bars I see are 8 pix wide.

[NY]

"Andy Furniss" <an...@andromeda64.plus.com> wrote in message
news:voCdnY5548C4F_bP...@brightview.co.uk...

I could have sworn that one broadcaster (maybe BBC) routinely used 704 and
another (maybe ITV) used 720. But when I check my recordings in my library,
all of them seem to be 720, irrespective of whether they are BBC, ITV, CH4
or one of the digital-only channels. Am I going mad?

Paul, in your derivation of the frequency, where does the factor 1000/1001
come from?

Is 2.2 a standard factor of safety over and above the theoretical rule from
Nyquist that sampling frequency must be at least 2x highest frequency in the
analogue signal - to allow for low-pass filters that have a non-cliff-edge
frequency response?

[Andy Furniss]

NY wrote:

> I could have sworn that one broadcaster (maybe BBC) routinely used
> 704 and another (maybe ITV) used 720. But when I check my recordings
> in my library, all of them seem to be 720, irrespective of whether
> they are BBC, ITV, CH4 or one of the digital-only channels. Am I
> going mad?

I've got a 2007 snippet from ITV1 DVB-T which is 704, I think BBC1
always used 720.

I don't know about now, but there was a time when ITV1 was 720 on DVB-T
and 704 on DVB-S - unless it was the other way round :-)

[Mark Carver]

NY wrote:

> I could have sworn that one broadcaster (maybe BBC) routinely used 704
> and another (maybe ITV) used 720. But when I check my recordings in my
> library, all of them seem to be 720, irrespective of whether they are
> BBC, ITV, CH4 or one of the digital-only channels. Am I going mad?

No. It used to be that BBC and C5 used 720, ITV and C4 704, that was DTT
I can't remember if the same applied to D-Sat or not.

[Paul Ratcliffe]

On Sat, 26 Oct 2013 11:45:47 +0100, NY <m...@privacy.net> wrote:

>> Having just looked at a recording from BBC2 last night, it does seem
>> that the BBC do, as mentioned in the thread, round up to 704 to be mpeg
>> friendly - the black bars I see are 8 pix wide.
>
> I could have sworn that one broadcaster (maybe BBC) routinely used 704 and
> another (maybe ITV) used 720. But when I check my recordings in my library,
> all of them seem to be 720, irrespective of whether they are BBC, ITV, CH4
> or one of the digital-only channels. Am I going mad?

You mean 720 of active picture on all, not 704 + 16 of black?

> Paul, in your derivation of the frequency, where does the factor 1000/1001
> come from?

Not too hot on NTSC, but I think it was a fudge factor introduced to reduce
correlated subcarrier dot crawl on luminance when colour (er, color) came in.
A similar thing happens in PAL, but it was achieved with a 25Hz offset.

> Is 2.2 a standard factor of safety over and above the theoretical rule from
> Nyquist that sampling frequency must be at least 2x highest frequency in the
> analogue signal - to allow for low-pass filters that have a non-cliff-edge
> frequency response?

Yes, basically (similar applies to CD at 20kHz and 44.1 sample freq.).
If the 5*2.25MHz multiple had been used it would have been 11.25MHz which
is only 2.045 times the video bandwidth.
The practicalities of designing such a filter without excessive passband
ripple and group delay problems would probably have been rather challenging.

[Paul Ratcliffe]

On Sat, 26 Oct 2013 12:59:33 GMT, Paul Ratcliffe
<ab...@orac12.clara34.co56.uk78> wrote:

>> Paul, in your derivation of the frequency, where does the factor 1000/1001
>> come from?
>
> Not too hot on NTSC, but I think it was a fudge factor introduced to reduce
> correlated subcarrier dot crawl on luminance when colour (er, color) came in.

Detail here:
http://en.wikipedia.org/wiki/NTSC

Seems it was to reduce beat frequencies between color and sound subcarriers
causing luminance dot crawl.

[NY]

"Paul Ratcliffe" <ab...@orac12.clara34.co56.uk78> wrote in message

news:slrnl6nfco...@news.pr.network...

Why did NTSC need to alter its frame rate when PAL simply needed to choose a
sensible sub-carrier frequency (ie n multiples of the line rate plus 25 Hz)?
Why couldn't the NTSC frequency be m multiples plus an offset? Does the
alternating nature PAL's carrier make the multiple-plus-offset approach
sufficient? Or was it because the US implementation of 525/30 happened to
have its sound offset from its vision by a spacing that made problems more
apparent than for the 625/25 systems used in Europe? I realise that
different European countries have different sound/vision spacing so what
works for mainland Europe may not work for UK/Ireland with a wider vision
bandwidth and bigger sound/vision spacing.

I presume that tweaking the frame rate was less of an issue than tweaking
the sound carrier spacing and hoping that the FM decoder in older TVs would
still be able to lock onto the sound signal.

It's just dawned on me that 1000/1001 is the factor that reduces 30 Hz to
29.97 Hz. I hadn't related the two :-(

[NY]

"Paul Ratcliffe" <ab...@orac12.clara34.co56.uk78> wrote in message

news:slrnl6nf5l...@news.pr.network...

> On Sat, 26 Oct 2013 11:45:47 +0100, NY <m...@privacy.net> wrote:
>
>>> Having just looked at a recording from BBC2 last night, it does seem
>>> that the BBC do, as mentioned in the thread, round up to 704 to be mpeg
>>> friendly - the black bars I see are 8 pix wide.
>>
>> I could have sworn that one broadcaster (maybe BBC) routinely used 704
>> and
>> another (maybe ITV) used 720. But when I check my recordings in my
>> library,
>> all of them seem to be 720, irrespective of whether they are BBC, ITV,
>> CH4
>> or one of the digital-only channels. Am I going mad?
>
> You mean 720 of active picture on all, not 704 + 16 of black?

I was meaning that the dimensions of the picture in the MPEG file that is
captured by DVB-T receivers and software (eg Windows Media Centre, XAMPP,
For The Record etc) is 704x576 for some/all recordings from ITV/CH4 and
720x576 (maybe a 704x576 picture in a wider black frame) for BBC, as
reported by MPEG editing/analysis programs such as VideoReDo, MediaInfo etc.


I've noticed that although programmes are broadcast with a full 576 lines of
picture, a few have a black half-line at the top-left and another at the
bottom-right - I thought that the need for half lines went out when analogue
equipment was phased out. I could understand it in archive recordings
originally made on analogue equipment, but some of it is modern programmes
which are presumably made/edited/transmitted in digital right from the
camera output to the TV receiver. Or do any broadcasters/studios still use
legacy PAL equipment?

[Jim Lesurf]

In article <slrnl6nf5l...@news.pr.network>, Paul Ratcliffe

<ab...@orac12.clara34.co56.uk78> wrote:
> On Sat, 26 Oct 2013 11:45:47 +0100, NY <m...@privacy.net> wrote:


> > Paul, in your derivation of the frequency, where does the factor
> > 1000/1001 come from?

> > Is 2.2 a standard factor of safety over and above the theoretical rule
> > from Nyquist that sampling frequency must be at least 2x highest
> > frequency in the analogue signal - to allow for low-pass filters that
> > have a non-cliff-edge frequency response?

> Yes, basically (similar applies to CD at 20kHz and 44.1 sample freq.).

I'm not quite clear what part of the comments on video you think "similarly
applies" to Audio CD. Can't say I recall any specific limit to 20kHz being
stated by Philips. Suspiciously rounded value. :-)

That said, a point which is often overlooked for this is that the sampling
theorem essentially tells us the sampled bandwith has to be *less* than
half the sample rate. So we'd certainly need to limit below 22.05kHz even
for absolutely perfect filters. 20kHz being a nice round number it may be
a value people adopt to quote/assume/aim at.

If you burrow down into the maths what emerges is that the limit is set by
the duration of the recording (or number of samples), again as a
theoretical limit. The longer the duration in scope of any filters, the
closer you can get to "22.05kHz" in theory.

More likely, though, any limit is better set by the behaviour of any
practical implimentation.

Slainte,

Jim

--
Electronics http://www.st-and.ac.uk/~www_pa/Scots_Guide/intro/electron.htm
Audio Misc http://www.audiomisc.co.uk/index.html
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html

[Mark Carver]

NY wrote:
> Or do any
> broadcasters/studios still use legacy PAL equipment?

I suspect there's still a bit of it in some of the 'yet to be modernised'
BBC regions. Plymouth, Newcastle, Southampton ?

The BBC's interview broom-cupboards in Reading and Brighton are still linked
to So'ton via BT CCTV grade PAL circuits for starters.

[NY]

"Jim Lesurf" <no...@audiomisc.co.uk> wrote in message
news:53a0c17...@audiomisc.co.uk...

> If you burrow down into the maths what emerges is that the limit is set by
> the duration of the recording (or number of samples), again as a
> theoretical limit. The longer the duration in scope of any filters, the
> closer you can get to "22.05kHz" in theory.

I'm not sure why the duration of the recording makes any difference to the
sharpness of cutoff of a 22.05 kHz LPF and therefore to the closeness that
you can approach the theoretical half-sampling-frequency limit. Maybe I'm
missing something very obvious...

When I studied electronic engineering I remember it being drummed into us
"thou shalt NEVER try to feed in signals that are greater than half the
sampling frequency or you'll incur the dreaded aliasing" (though I was
always disappointed that we never got to hear what aliasing sounded like).
And the way to achieve this was to use as sharp an LPF as possible and/or to
make sure that the sampling frequency was "comfortably" greater than the
cut-off frequency.

However I don't think "comfortably" was ever quantified - how *much* the
factor of safety should be. That's why I was interested to see a specific
figure of 2.2 mentioned as if it was a widely-acknowledged value.

[Roderick Stewart]

On Sat, 26 Oct 2013 15:42:18 +0100, "NY" <m...@privacy.net> wrote:

>Why did NTSC need to alter its frame rate when PAL simply needed to choose a
>sensible sub-carrier frequency (ie n multiples of the line rate plus 25 Hz)?

Different problems, different solutions.

The PAL 25Hz subcarrier offset was to create a relationship between
frame and subcarrier frequencies such that positive half cycles would
line up with negative ones on alternate frames. It was nothing to do
with sound or transmission.

This relationship was already correct in NTSC (just the way the
numbers worked out), so to offset the subcarrier for least visible
beat patterns with the sound carrier, it was necessary to offset
everything else as well. They could have offset the sound carrier
instead, but this could have caused problems with millions of existing
monochrome TV sets.

Rod.

[NY]

"Roderick Stewart" <rj...@escapetime.myzen.co.uk> wrote in message
news:oiho691k544s7h9m1...@4ax.com...

So was it just good design and planning that the sound-vision spacing was
already large enough for both European and UK PAL broadcast systems, or was
there more to it that this?

Was it due to the choice of colour subcarrier frequency - would making it a
different multiple of the line rate have helped to avoid interference from
sound?

[NY]

"Roderick Stewart" <rj...@escapetime.myzen.co.uk> wrote in message
news:oiho691k544s7h9m1...@4ax.com...

> On Sat, 26 Oct 2013 15:42:18 +0100, "NY" <m...@privacy.net> wrote:
>
>>Why did NTSC need to alter its frame rate when PAL simply needed to choose
>>a
>>sensible sub-carrier frequency (ie n multiples of the line rate plus 25
>>Hz)?
>
> Different problems, different solutions.
>
> The PAL 25Hz subcarrier offset was to create a relationship between
> frame and subcarrier frequencies such that positive half cycles would
> line up with negative ones on alternate frames. It was nothing to do
> with sound or transmission.
>
> This relationship was already correct in NTSC (just the way the
> numbers worked out)

Wasn't it the fact that PAL alternates the phase of the subcarrier whereas
NTSC doesn't that required the field-rate shift in subcarrier to make
positive cycles line up with negative ones on alternate frames?

[Jim Lesurf]

In article <34qdnQnm2u4gkPHP...@brightview.co.uk>, NY

<m...@privacy.net> wrote:
> "Jim Lesurf" <no...@audiomisc.co.uk> wrote in message
> news:53a0c17...@audiomisc.co.uk...
> > If you burrow down into the maths what emerges is that the limit is
> > set by the duration of the recording (or number of samples), again as
> > a theoretical limit. The longer the duration in scope of any filters,
> > the closer you can get to "22.05kHz" in theory.

> I'm not sure why the duration of the recording makes any difference to
> the sharpness of cutoff of a 22.05 kHz LPF and therefore to the
> closeness that you can approach the theoretical half-sampling-frequency
> limit. Maybe I'm missing something very obvious...

It isn't obvious, so you're not alone in not having been aware of this
detail! And it says nothing explicit about filters as such. Just about the
range of frequencies (bandwidth) of the signal which will be OK in theory.

The maths/logical argument that explains it is bit of a journey. And in
practice you often get a value so close to half the sampling frequency that
no-one cares anyway. So they just say "a little less than half the sample
rate". :-)

In essence, though, it stems from the imprecision of 'frequency' in FTs on
a finite set of sampled values. In practice the resolution for general
waveforms where the samples are the *only* info you have mean that any
signal 'close enough' to fs/2 becomes indistinguishable from fs/2 *unless*
you have other information (e.g. being told the signal *is*, say, a pure
sinewave.)

The resolution is set for such general waveforms by the number of samples /
duration of the set of values. So in effect you can explain it by saying
that for, say, a sample series of 1 second duration the resolution is about
1 Hz. So you'd then be OK with inputs up to 1 Hz below fs/2.

In practice, of course, you'd start any real practical cut-off below that.
And probably only have a shorter set of values in scope of an FIR digital
filter which is the most common method in audio as it is the easiest to
design and apply.

More formally, it ends up being linked to various assumptions and details.
e.g. some aspects assuming 2N samples are needed, others needing 2N+1
when doing processes.

In practice the result of overlooking this and assuming you can go right up
to and allow fs/2 is 'pathalogical' situations such as an input sinewave at
fs/2 giving a series of zeros as samples if its phase happens to place its
zero crossings at the sampled instants. Nature's way of warning you that
you are not allowed to do this and are pushing yer luck! 8-]

At the other end of the spectrum it also leads to the point that all the
frequencies in a spectrum require two values - e.g. amplitude and phase -
to be defined. But that dc has no 'phase' variable in any real sense.

FWIW It is also common for people to assume that the limit is on the
highest frequency allowed. But in fact the Sampling Theorem is about
*bandwidth*.

[Roderick Stewart]

On Sun, 27 Oct 2013 09:13:47 -0000, "NY" <m...@privacy.net> wrote:

>>
>>>Why did NTSC need to alter its frame rate when PAL simply needed to choose
>>>a
>>>sensible sub-carrier frequency (ie n multiples of the line rate plus 25
>>>Hz)?
>>
>> Different problems, different solutions.
>>
>> The PAL 25Hz subcarrier offset was to create a relationship between
>> frame and subcarrier frequencies such that positive half cycles would
>> line up with negative ones on alternate frames. It was nothing to do
>> with sound or transmission.
>>
>> This relationship was already correct in NTSC (just the way the
>> numbers worked out), so to offset the subcarrier for least visible
>> beat patterns with the sound carrier, it was necessary to offset
>> everything else as well. They could have offset the sound carrier
>> instead, but this could have caused problems with millions of existing
>> monochrome TV sets.
>
>So was it just good design and planning that the sound-vision spacing was
>already large enough for both European and UK PAL broadcast systems, or was
>there more to it that this?
>
>Was it due to the choice of colour subcarrier frequency - would making it a
>different multiple of the line rate have helped to avoid interference from
>sound?

Not sound/vision spacing, sound/subcarrier spacing. I don't think it
was a matter of how large the spacing was, but the numerical
relationship between the beat pattern and the scanning frequencies.
Like the subcarrier frequency itself, it had to be an optimum value
for minimum visibility.

There may have been a different multiple of line rate that would have
worked, but all the interrelationships are such that if you change
anything you have to change something else as well. A colleague of
mine once worked out a system that would have obviated the need for
the 25Hz offset in PAL, and the consequent 8 field editing sequence,
but it would have required changing to 627 lines. I can't remember
whether it was the subcarrier frequency or the field scanning rate
that was maintained, as it was a long time ago, but one of them would
have had to change.

The PAL subcarrier polarity reversals are effectively just a 7.8kHz
modulation, so don't really affect any of this.

It really is a shame that television wasn't in colour from the very
start, as we could perhaps have designed a coherent system that was a
lot simpler. The trouble with inventing something as a modification is
that it has to fit with what's already there. You can't just scrap
millions of TVs and start again.

Rod.

[michael...@live.co.uk]

In simple terms NTSC isn't 'm' multiples of line rate. NTSC has a 'half line' offset of an odd multiple of the NTSC line frequency. This means the peaks and troughs that Roderick has previously mentioned on alternate lines don't align on adjacent TV lines. It also means the NTSC subcarrier sidebands neatly interleave with the luminance freq components. This wouldn't work with PAL because the alternate phase of the subcarrier freq would cause these peaks and troughs to line up on alternate lines.(UK) PAL subcarrier is an odd multiple of a quarter freq of line frequency to achieve the same effect. But additionally a 25hz offset was added to ensure a phase reversal on each successive field to make the peaks and troughs even further apart. The quarter line offset also ensures that the sidebands of the subcarrier neatly interleave with the luminance signal on every line. If (UK) Pal was simply a multiple of a half line odd multiple of the line freq, the sub carrier and luminance frequency components would interfere with each other on every other line.

I once saw a very effective demonstration of this when I was studying tv engineering. The lecturer had a Tek spg that you could switch the subcarrier 25hz offset off. Looking at 'normal' 100% colour bars with the 25 hz switched on gave the usual picture engineers are used to seeing with regard to the perception of the 'chroma dots'. When you switched the 25hz offset 'off' the chroma dots were very much more apparent to the eye - they 'moved' much more slowly. From memory this was demonstrated on a colour monitor with no filtering in the luminance path. So you could argue the 25hz offset was a nicety rather than a necessity.

[Roderick Stewart]

On Sun, 27 Oct 2013 06:39:27 -0700 (PDT), michael...@live.co.uk
wrote:

>
>I once saw a very effective demonstration of this when I
>was studying tv engineering. The lecturer had a Tek spg that
>you could switch the subcarrier 25hz offset off. Looking at
>'normal' 100% colour bars with the 25 hz switched on gave
>the usual picture engineers are used to seeing with regard to
>the perception of the 'chroma dots'. When you switched
>the 25hz offset 'off' the chroma dots were very much more
>apparent to the eye - they 'moved' much more slowly.
>From memory this was demonstrated on a colour monitor
>with no filtering in the luminance path. So you could argue
>the 25hz offset was a nicety rather than a necessity.

Don't forget heterodyne colour systems like VHS, where the output
subcarrier was completely unrelated to syncs, or even "random
interlace" systems used in some cheap security cameras where even the
line and field syncs were derived from separate free-running
oscillators not linked to each other at all. Compared with broadcast
quality, these were rubbish, but they worked. You can get away with
murder as long as the result doesn't have to be broadcast, or be
guaranteed to work with anybody else's equipment every time.

Rod.

[NY]

"Roderick Stewart" <rj...@escapetime.myzen.co.uk> wrote in message

news:83bq691nckfd8tq7f...@4ax.com...

Yes I've seen all sorts of non-standard pseudo-PAL outputs from
non-broadcast devices. One of my digital cameras that could produce a PAL
output to display the photos on a TV worked fine on a newer TV from 2000,
but the colours flickered and there were pronounced Hannover bars and
tearing of the top and bottom of the picture on an older TV from the late
1980s. Trying to record the picture on VHS (I was playing, to see what did
and didn't work!) produced some painful graunching noises from the VCR as
presumably the servos tried and failed to lock onto the signal.

Then there's the hybrid NTSC/PAL output that some newer VHS VCRs can produce
when playing an NTSC tape. I believe this is NTSC line/frame rate but with
PAL colour sub-carrier. I'm not sure whether it's 4.43 without alternating
phase or whether it's 3.58 (NTSC). Again, the newer TV could display that
pretty well (stable picture and colours) but the older one couldn't sync
with the line/frame rate and consequently it was very difficult to tell
whether the colours flickered.

These days you forget about analogue anomalies like cross-colour and
subcarrier-dot-patterning. I bet there are many young people who will never
have seen them unless they happen to watch programmes from the archive made
on PAL equipment.

The 4-frame edit window of PAL must have imposed a lot of creative
limitations, given that editors could only edit to an accuracy of the
nearest 4/25 of a second, whereas film can be edited to the nearest 1/25
sec.

Recently I was cleaning up a short film that a group of us made on an
"understanding video" night-school course about 20 years ago. I think it was
originally edited on domestic VHS equipment (I don't think the course leader
copied the rushes to U-Matic and edited on that because he said it would
have been cheating - we had to use equipment that was available to anyone
rather than just to professionals), and there was a nasty flicker on the
colour at every edit, possibly due to contravening the 4-frame rule. Having
digitised it to MPEG, it was very easy using VideoReDo to take out one frame
at each edit to hide that effect, though it did become tedious after I'd
done it a few times! Luckily there were very few edits where there was
dialogue at the time, apart from one insert edit where he dropped in a mute
close-up over part of a longer long shot, keeping the sound from the long
shot, and I decided to leave those two edits alone.

Likewise for tidying up an MPEG produced from (silent) 8 mm home cine films,
where it was necessary to take out the very noticeable jagged line wherever
the film had been spliced.

One day, as a class exercise, I'll take a commercially-produced DVD of
amateur movies of old trains, with a commentary and some sections with
lip-sync pieces to camera, and see if I can tidy up the splices. Chopping
out the film splices from the pictures is easy but it will disturb the sound
track so I'll need to copy that to a separate WAV file and make sure that
for every frame of film I chop out, I chop out an equivalent 1/25 second
from the soundtrack somewhere nearby that's not noticeable, so as to keep
sound and vision in sync for the places where it matters - the modern
lip-sync sections. And then merge the pictures and sound back together. It
was rather amateurish that the producer of the DVD hadn't tidied up the film
splices before adding the commentary.

[Roderick Stewart]

On Sun, 27 Oct 2013 16:25:00 -0000, "NY" <m...@privacy.net> wrote:

>Then there's the hybrid NTSC/PAL output that some newer VHS VCRs can produce
>when playing an NTSC tape. I believe this is NTSC line/frame rate but with
>PAL colour sub-carrier. I'm not sure whether it's 4.43 without alternating
>phase or whether it's 3.58 (NTSC).

Oh yes, "NTSC443" is what that was called. A completely non-standard
"standard" but widely used because it enabled a cheap way to play back
foreign video without properly transcoding it. Since the off-tape
frequency was changed by simply heterodyning it, without any other
complications, the output from an NTSC tape would still be NTSC
without the PAL phase alternation. The decoder in the monitor could be
made to work both standards with much of the same circuitry because
both used the same frequency, the PAL switching circuitry being
disabled when not required. Presumably this was cheaper than including
a completely separate decoder in the monitor with crystals and tuned
circuits for 3.58MHz

I'm not sure if there was ever a corresponding "PAL358" system to
allow the Yanks to play European tapes, but theoretically there could
have been, except it would have required a bit of extra circuitry they
wouldn't normally need, for the PAL switching, so maybe the cost
saving wouldn't have been so great.

Rod.

[NY]

"Roderick Stewart" <rj...@escapetime.myzen.co.uk> wrote in message

news:j6pq69hhstqbe9jkj...@4ax.com...

My understanding is that multi-standard TVs/VCRs capable of playing PAL
broadcasts/tapes in the American market are *very* rare, whereas most recent
TVs/VCRs sold in the UK (and maybe the rest of Europe) can play NTSC. When
my sister was living in the US for a couple of years, she had to hunt far
and wide for TV/VCR and it was crap compared with dedicated PAL equipment in
that it was far more fussy as to whether it could play multi-generation or
non-broadcast material.

Wouldn't a PAL358 system still need to have phase-reversing circuitry to
reverse the phase of the alternate-phase lines from a PAL tape to put all
lines back into the same phase as needed by an NTSC TV?

[J. P. Gilliver (John)]

In message <2avp69dmgfc8e42ld...@4ax.com>, Roderick
Stewart <rj...@escapetime.myzen.co.uk> writes:
[]

>It really is a shame that television wasn't in colour from the very
>start, as we could perhaps have designed a coherent system that was a
>lot simpler. The trouble with inventing something as a modification is
>that it has to fit with what's already there. You can't just scrap
>millions of TVs and start again.
>
>Rod.

When TVs cost a higher proportion of people's income, yes. They've had
few qualms about making lots of set-top boxes scrap in the relatively
short time FreeView's been going, and there are several frequent posters
here who want to do the same for DAB.
--
J. P. Gilliver. UMRA: 1960/<1985 MB++G()AL-IS-Ch++(p)Ar@T+H+Sh0!:`)DNAf

Santa's elves are just a bunch of subordinate Clauses.

[SpamTrapSeeSig]

In article <53a122c...@audiomisc.co.uk>, Jim Lesurf
<no...@audiomisc.co.uk> writes

>In essence, though, it stems from the imprecision of 'frequency' in FTs on
>a finite set of sampled values. In practice the resolution for general
>waveforms where the samples are the *only* info you have mean that any
>signal 'close enough' to fs/2 becomes indistinguishable from fs/2 *unless*
>you have other information (e.g. being told the signal *is*, say, a pure
>sinewave.)

Jim, you probably know me well enough via this group to know my
appreciation of the maths isn't what it ought to be, but anyway...

... I'm puzzling over this:

I can appreciate theoretically that Nyquist requires fs/2 -x, and that
the recording duration will have a bearing, in that it determines the
number of samples available in toto ('x' should tend to zero for very
long recordings).

Surely though, by definition, a signal at the chosen limit of bandwidth
must be regarded as 'sinusoidal'? Where do FTs come into this? Are you
assuming sampling before digital 'filtering' using Fourier?

S.
(who was probably texting surreptitiously under the desk at the crucial
moment)

--
SimonM

[Roderick Stewart]

On Sun, 27 Oct 2013 20:38:13 -0000, "NY" <m...@privacy.net> wrote:

>
>Wouldn't a PAL358 system still need to have phase-reversing circuitry to
>reverse the phase of the alternate-phase lines from a PAL tape to put all
>lines back into the same phase as needed by an NTSC TV?

Yes it would, which is why I don't think it was widely used, if at
all. An American dual-standard decoder handling both normal NTSC358
and "PAL358" (if it existed) would have to include PAL switching
circuitry just for the rare occasion when it was presented with replay
of a European tape. Most of the time the extra circuitry wouldn't be
used, if ever, so most manufacturers would see no reason for the extra
expense. Any American mad enough to want to play a European PAL tape
could either obtain some European equipment, or a standards converter.

In PAL countries, the extra circuitry has to be there anyway, so it's
no big deal to disable it automatically when not required, as long as
the rest of the decoder doesn't have to change frequency as well.
Playing NTSC and PAL tapes with the same output subcarrier frequency
makes the VHS player circuitry simpler anyway. In other words,
decoding NTSC443 is a very useful extra feature for minimal cost, and
I've found it present in many PAL TV sets and monitors even when not
mentioned in the spec.

Rod.

[Jim Lesurf]

In article <ojorHDRE...@virginmedia.net>, SpamTrapSeeSig

<no-...@nospam.virginmedia.net> wrote:
> In article <53a122c...@audiomisc.co.uk>, Jim Lesurf
> <no...@audiomisc.co.uk> writes
> >In essence, though, it stems from the imprecision of 'frequency' in FTs
> >on a finite set of sampled values. In practice the resolution for
> >general waveforms where the samples are the *only* info you have mean
> >that any signal 'close enough' to fs/2 becomes indistinguishable from
> >fs/2 *unless* you have other information (e.g. being told the signal
> >*is*, say, a pure sinewave.)

> Jim, you probably know me well enough via this group to know my
> appreciation of the maths isn't what it ought to be, but anyway...

> ... I'm puzzling over this:

> I can appreciate theoretically that Nyquist requires fs/2 -x, and that
> the recording duration will have a bearing, in that it determines the
> number of samples available in toto ('x' should tend to zero for very
> long recordings).

Yes. Its usually 'small' and tends to zero as the duration (in scope) gets
bigger.

> Surely though, by definition, a signal at the chosen limit of bandwidth
> must be regarded as 'sinusoidal'? Where do FTs come into this? Are you
> assuming sampling before digital 'filtering' using Fourier?

The problem can be described in logical/philosopical terms although you end
up needing the maths to quantify the actual effect.

The presumption is that once sampling has been done the *only* information
you have is the series of sample values plus some very simple 'recipy'
details like "Samples were taken 1/44100th of a second apart". Plus the
assurance that it is a "complete record".

This means that having been given the samples you have no other info on
what the actual waveform shape was that was sampled. You don't "know" what
waveform shape was "chosen" to be sampled. You have to reconstruct it using
the correct method, just using the sample values.

This situation is quite different from being *also* told something like
"The input waveform *was* a pure sinewave".

All real FTs on real data have finite durations (and hence numbers of
samples for sampled FTs) and contain a finite amount of information.

Have a look at the start of chapter/lecture 7 from

http://www.st-andrews.ac.uk/~www_pa/Scots_Guide/iandm/intro.html

Ignore the maths and look at the diagrams. One a aspect of the above is
that to do an FT we have to decide what form of FT to use. In general we
presume a waveform which we assume was periodic with a period equal to the
sampled duration (call that T). This lets us choose a finite set of
frequencies for our computed spectrum. We can then say the waveform had an
amount at a frequency 1/T, some other amount at 2/T, etc. These are then
the *only* frequencies we can presume are present. We have no value for the
amount at "just above 2/T". But the set of amplitudes and phases for these
frequencies completely defines the sampled waveform. Any contributions at
"other frequencies" shows up via these chosen ones in just the correct way
to get the right waveform.

Under this is another point that most books pass over and many people don't
realise. That there is also more than one definition of "frequency" and
that the word is used in different ways for different purposes in various
circumstances.

The upshot of all that for the question of how close we can get to fs/2 is
that we can't resolve once we get too close to fs/2 because of the way we
have "defined the term frequency" for our FT processes.

Now if we *know* that the signal was, say, a pure sinesoid we could do
better. Indeed, in that case only a few samples can be used to work out a
very precise frequency, amplitude, and phase for that sinwave.

Lack of knowing about these details is very common even amongst experienced
academics and engineers. I once had a grant refused because I was told what
I proposed was "impossible" as a result.

I took the refusal letter down to the lab, held it up to an instrument I
already had working and told the instrument what it was doing was
"impossible". Alas, being very naughty, it didn't bow down to the expert
opinion and just went on doing the "impossible". :-)

The instrument was a Martin-Puplett (polarising Michaelson) interferometer.
Nomally these are used with general FFTs to do spectra, etc. That works as
per the usual FT assumptions. However I was using it to measure the
oscillation frequency of stable (Gunn) oscillators. These would radiate a
very clean signal in the 100GHz region.

Since we knew they were very periodic at a 'frequency' somewhere around
100GHz we could count the spacings between zero crossings of their
wavelengths using the interfereometer and measure the frequency to about
10MHz. Which was orders of magnitude better than a conventional FFT
approach. Hence "impossible", despite working very nicely. :-)

On more than one occasion I had grants rejected on a basis that a learned
professor on a board thought the proposal was "impossible". Quite often I'd
already built a working prototype to check it worked, and wanted the money
to make a better-constructed and more reliable version. [1]

In the end I got most of my R&D money from companies and outside bodies.
They were more interested in the fact that the instruments worked than that
they were "impossible". ;->

Jim

[1] Yet I've also seen many absurd ideas given large grants, and duly fail.

[Paul Ratcliffe]

On Sat, 26 Oct 2013 17:36:32 +0100, Mark Carver <mark....@invalid.invalid>

wrote:

>> Or do any broadcasters/studios still use legacy PAL equipment?
>
> I suspect there's still a bit of it in some of the 'yet to be modernised'
> BBC regions. Plymouth, Newcastle, Southampton ?

Southampton was digitised years ago. Can't imagine Newcastle isn't, but
I don't know. Plymouth's about to be refurbed.

> The BBC's interview broom-cupboards in Reading and Brighton are still linked
> to So'ton via BT CCTV grade PAL circuits for starters.

So are our four outstations.

[Paul Ratcliffe]

On Sat, 26 Oct 2013 17:19:33 +0100, Jim Lesurf <no...@audiomisc.co.uk> wrote:

>> Yes, basically (similar applies to CD at 20kHz and 44.1 sample freq.).
>
> I'm not quite clear what part of the comments on video you think "similarly
> applies" to Audio CD. Can't say I recall any specific limit to 20kHz being
> stated by Philips. Suspiciously rounded value. :-)

I don't understand what you don't understand (minister). Sampling applies
to both audio and video and the same rules apply regarding Nyquist.
It may not be a chapter-and-verse spec., but CD is always quoted as 20kHz
whenever I've looked at manufacturers' manuals etc.

You can pontificate and post as much theory as you like and take pages
doing it. I can state that a good 'rule of thumb' is 2.2 in the real
world, and do it in one line.

> That said, a point which is often overlooked for this is that the sampling
> theorem essentially tells us the sampled bandwith has to be *less* than
> half the sample rate. So we'd certainly need to limit below 22.05kHz even
> for absolutely perfect filters.

You can't build an absolutely perfect filter, as you well know, as it would
have to start outputting before it had received any input.

> If you burrow down into the maths what emerges is that the limit is set by
> the duration of the recording (or number of samples), again as a
> theoretical limit. The longer the duration in scope of any filters, the
> closer you can get to "22.05kHz" in theory.

How long is a recording then? As long as that piece of string? Who cares
how close you can get to any particular magic number? It's always a law
of diminishing returns.

> More likely, though, any limit is better set by the behaviour of any
> practical implimentation.

Hooray!

[Paul Ratcliffe]

On Sat, 26 Oct 2013 15:51:33 +0100, NY <m...@privacy.net> wrote:

> I've noticed that although programmes are broadcast with a full 576 lines of
> picture, a few have a black half-line at the top-left and another at the
> bottom-right - I thought that the need for half lines went out when analogue
> equipment was phased out.

It did, although analogue PAL hasn't been phased (!) out for several uses
in studio centres. I don't know why some things appear to have analogue
type blanking applied to signals which are and always have been in the
digital domain.

[Paul Ratcliffe]

On Sun, 27 Oct 2013 06:39:27 -0700 (PDT), michael...@live.co.uk

<michael...@live.co.uk> wrote:

[Do you have to post massive long lines without any line breaks?
It's very irritating to read (and reply to).]

> But additionally a 25hz offset was added to ensure a phase reversal on
> each successive field to make the peaks and troughs even further apart.
> The quarter line offset also ensures that the sidebands of the
> subcarrier neatly interleave with the luminance signal on every line.

Of course these two facts are linked. You don't get one effect without
the other - they are indeed the same effect described differently.

[Paul Ratcliffe]

On Sun, 27 Oct 2013 19:28:29 +0000, Roderick Stewart
<rj...@escapetime.myzen.co.uk> wrote:

> I'm not sure if there was ever a corresponding "PAL358" system to
> allow the Yanks to play European tapes, but theoretically there could
> have been

Most Yanks wouldn't know there was any other system in the first place.
Most of 'em probably think Europe is another part of the USA.
If not, then why would they want to play something from there anyway?
"If it's worth watching, it'll be in NTSC" attitude.

[NY]

"Paul Ratcliffe" <ab...@orac12.clara34.co56.uk78> wrote in message

news:slrnl6sigl...@news.pr.network...

What is weird is that the initial black line occupies about 2/3 of the
screen width rather than half the width as it would with a PAL half-line:

http://img443.imageshack.us/img443/3388/uk7v.png

(The MPEG display software, VideoReDo, has rescaled the 704x576 MPEG to
1024x768 when it captured the still, but the principle still applies - the
black line extends to about pixel 666 which is 65% of the width.)

This is from ITV and presumably the programme is made all in the digital
domain. The effect seems to be confined to ITV and CH4. CH5 has a full-width
but blurred top line; the various BBC channels are full 576 lines with no
initial black part-lines.

[Andy Furniss]

Seems the BBC used to (or maybe it varies).

Here's an unscaled old shot from the BBC - I see the lines are inverted
compared to yours with short on top and long on bottom.

http://imageshack.us/a/img39/4763/0i.png

[Jim Lesurf]

In article <slrnl6si4p...@news.pr.network>, Paul Ratcliffe

<ab...@orac12.clara34.co56.uk78> wrote:
> On Sat, 26 Oct 2013 17:19:33 +0100, Jim Lesurf <no...@audiomisc.co.uk>
> wrote:

> >> Yes, basically (similar applies to CD at 20kHz and 44.1 sample freq.).
> >
> > I'm not quite clear what part of the comments on video you think
> > "similarly applies" to Audio CD. Can't say I recall any specific limit
> > to 20kHz being stated by Philips. Suspiciously rounded value. :-)

> I don't understand what you don't understand (minister).

Your reply does show that to be true. :-)

> Sampling applies to both audio and video and the same rules apply
> regarding Nyquist. It may not be a chapter-and-verse spec., but CD is
> always quoted as 20kHz whenever I've looked at manufacturers' manuals
> etc.

You may be confusing this with a common for makers in manuals to make a
statement like "Frequency response: 20Hz - 20kHz +/- XdB". This doesn't
come from 20kHz being 10 percent less than 22.05kHz. It comes from that
range being adopted as useful as the normal 'hearing range' for specs of
spectral flatness - e.g. also in amplifiers.

> You can pontificate and post as much theory as you like and take pages
> doing it. I can state that a good 'rule of thumb' is 2.2 in the real
> world, and do it in one line.

Yes, you can state whatever you choose. :-) It may help if you notice that
I haven't made any comment on your assertion wrt *video*.

However I've been commenting on other matters brought up by other people.
Don't worry, though, no-one is demanding that you understand that
discussion. :-)

> > That said, a point which is often overlooked for this is that the
> > sampling theorem essentially tells us the sampled bandwith has to be
> > *less* than half the sample rate. So we'd certainly need to limit
> > below 22.05kHz even for absolutely perfect filters.

> You can't build an absolutely perfect filter, as you well know,

Indeed.

> as it would have to start outputting before it had received any input.

Erm, no, that isn't quite the reason. But if you haven't followed what I've
already written I won't explain further.

> > If you burrow down into the maths what emerges is that the limit is
> > set by the duration of the recording (or number of samples), again as
> > a theoretical limit. The longer the duration in scope of any filters,
> > the closer you can get to "22.05kHz" in theory.

> How long is a recording then? As long as that piece of string? Who cares
> how close you can get to any particular magic number?

Those I've been talking to in response to their comments/questions. If that
doesn't interest you or passes over your head, no-one else is requiring you
to join in.