Re: 1080i & 720p HDTV Resolution

From: DAB sounds worse than FM (dab_is_at_low.quality)
Date: 02/15/05 

Date: Tue, 15 Feb 2005 15:07:20 GMT

Kevin Bracey wrote:
> In message <Oy7Qd.147$Zp3.50@newsfe2-gui.ntli.net>
> "DAB sounds worse than FM" <dab_is@low.quality> wrote:
>
>> Kevin Bracey wrote:
>>> In message <qX4Qd.198$YO3.155@newsfe4-gui.ntli.net>
>>> "DAB sounds worse than FM" <dab_is@low.quality> wrote:
>>>
>>>> Basically, it's all very well giving these extreme examples, but in
>>>> reality when do you ever see alternating black and white lines?
>>>
>>> Just had another thought on this - thanks to William Snyder's reply
>>> to you in sci.image.processing.
>>>
>>> He also separates twitter into two parts - the vertical judder
>>> effect, and the black/white line problem, which he points out is
>>> just that of aliasing - a concept you should be familiar with in
>>> audio. We all now agree that the visibility of twitter/judder
>>> diminishes with higher resolution. But as William says, the
>>> aliasing problem does not. The details on someone's shirt that used
>>> to alias at 576i no longer will, but with HDTV now even finer
>>> details have become visible that could alias at 1080i.
>>
>>
>> Yes, but he also says this:
>>
>> "the twitter part may be somewhat less visible because to the viewer
>> of 1080i the scanlines are closer in angular frequency (all else
>> being equal) so are filtered more by the human optical system"
>
> You're suffering reading comprehension failure again. He's separated
> the phenomena into two parts the twitter (ie vertical judder) and the
> aliasing problem. There he is talker the twitter/judder, which you, I
> and he all agree is diminished by more lines. Read the paragraph I
> just wrote again. He is not talking about the aliasing.

If he's not talking about aliasing then why does he write the
following?:

"i haven't read the book but "twitter" probably refers to the visible
interaction between aliasing and interlacing"

Twitter IS due to vertical aliasing.

>> so the finer detail is less likely to be able to be resolved by the
>> human optical system because it's at a higher spatial frequency.
>> Think about the alternating black and white lines which merge to
>> grey.
>
> For God's sake, Steve, can't you understand this? The alternating
> black and white lines are temporally separated by the interlacing
> process to appear on alternate fields. IN THE RESULTING SIGNAL THEY
> NEVER APPEAR AT THE SAME TIME, SO THE ABILITY OF THE HUMAN OPTICAL
> SYSTEM TO RESOLVE THEM HAS NOTHING TO DO WITH IT. IT'S ABOUT THE
> ODD/EVEN LINE SAMPLING PERFORMED BY THE INTERLACING.

Calm down dear.

AIUI, progressive displays display all lines at the refresh rate. So,
assuming no motion of these hypothetical alternating black and white
lines, the progressive display should display alternating black and
white lines every 20ms. Then again, they won't be black and white
anyway, because there's a vertical lowpass filter.

> Regardless of
> whether your interlaced system has 20 lines, 576 lines, 1080 lines or
> 4096 lines, an horizontal bar pattern with frequency of 1/2 the line
> frequency will be transformed by the interlacing into a flat 25Hz
> strobe effect.

I think you're talking about interlaced displays.

> And even a progressive display showing such a signal will have to
> still show the picture as a 25Hz strobe, as that is what the signal
> contains.

Say the progressive display receives the (even-numbered lines) field
with all white lines, then it'll update all the even-numbered lines to
be white. But all the odd numbered lines will be black. It should then
display the alternate black and white lines.

> If the alternate fields are black and white, then that's
> what it will display. It couldn't reconstruct it back into a
> stationary black/white line pattern unless it knew (via auxiliary
> signalling or whatever) that it is actually a film mode source.
>
>> But because the human optical system is effectively a lowpass filter,
>> then halving the line-pitch pushes the fine detail up in spatial
>> (angular?) frequency, and your eyes cannot resolve things at such
>> high frequencies.
>
> So, then, why doesn't the argument apply to audio? Why don't you
> argue that the cut-off for 44kHz audio should be higher than 22kHz
> because the ear can't hear >22kHz anyway.

Firstly, if you have aliased high audio frequencies down to low
frequencies then that could ruin the entire sound. With digital TV, if
you have some aliasing then it could be limited to as little as one
16x16 pixel macroblock, and you might not even perceive the problem, and
it certainly does not necessarily ruin the entire picture.

Secondly, the bandwidth for audio is set by the frequency response of
the typical / good human auditory system (i.e. up to about 20kHz tops).
I don't know this, but I'd guess it's correct: different musical
instruments probably have harmonics with energy well above 20kHz, so
such higher frequencies would be folded down to lower frequencies and
screwing up the audio. To quote Poynton: "typical images are dominated
by power at low spatial frequencies," and I think that the amplitudes
quickly tail-off as spatial frequency increases. Poynton shows the
output of a typical DCT on page 449, where low spatial frequencies
certainly dominate, and higher spatial frequencies are almost
non-existent, and would very likely be removed in the quantisation
process.

When you alias a spectrum, then you fold down the spectrum. For example,
if you have a time signal with Fs=10kHz and a sinusoid with f=7.5kHz,
then the alias will be at Fs - f = 2.5kHz. So, if you allow some
aliasing for a macroblock, then because the higher spatial frequencies
tend to have lower amplitudes, then you're folding-down low amplitudes
to lower frequencies, and it usually won't be a problem.

Remember that all DTT TV you watch uses a vertical lowpass filter that
allows aliasing. But CDs and digital radio stations do obey Nyquist's
criterion.

> Think about it. It's not
> about the human hearing the 22kHz signals, it's about the >22kHz
> signals mucking up the sampling and creating audible artifacts below
> 22kHz. Just like the video - it's not about the humans seeing the
> high frequency vertical information, it's the high frequency vertical
> information mucking up the interlacing and causing temporal effects.

See above.

> Now, in another post you seem to be arguing that you never see
> black/white lines anyway because of the filtering. (!) Well, yes, and
> the nature of that filtering is what we're discussing here. You're
> arguing that that filtering can be further reduced. And I'm just
> trying to point out that that will make the problem of black/white
> line patterns worse.

I've never denied that it would make the black-white lines pattern
worse. All I'm saying is that because the line-pitch is about half what
it is for SDTV then it should be less perceptible EVEN if the lowpass
filter cutoff was relaxed. 

-- 
Steve - www.digitalradiotech.co.uk - Digital Radio News & Info
Find the cheapest Freeview, DAB & MP3 Player Prices:
http://www.digitalradiotech.co.uk/freeview_receivers.htm
http://www.digitalradiotech.co.uk/dab_digital_radios.htm
http://www.digitalradiotech.co.uk/mp3_players_1GB-5GB.htm
http://www.digitalradiotech.co.uk/mp3_players_large_capacity.htm

Relevant Pages

  • Re: Image rescaling algorithms
    ... completely the aliasing isn't a task that an mere mortal can do... ... - Filter the image to remove high frequencies; ... obtain an high performance filter a box filter computed using the integral ...
    (sci.image.processing)
  • Re: Filter design.
    ... The amount of the distortion due to aliasing equals to the sum of power ... The first order filter is just the first order filter. ... So I HAVE to sample at 8kHz and sadly I do not have very good analog pre-filtering for this sampling rate. ... at frequencies higher than 4kHz will cause aliasing after sampling. ...
    (comp.dsp)
  • Re: New Coolscans when?
    ... frequencies are still high and so not real noticeable. ... The real bad aliasing is when you fold all the way ... to be this sharp on a pixel level and not have some aliasing. ... scanner can over resolve and perhaps see a bit of grain aliasing. ...
    (comp.periphs.scanners)
  • Re: a mini question about the "sample sheme" of OFDM
    ... Keep in mind that frequencies that appear to ... be beyond the Nyquist frequency will end up wrapping (aliasing) around ...
    (comp.dsp)
  • Re: Earl & Donna Evleths address
    ... >> Spam filters are quite good, our Wanadoo filter gets about 90% of it ... But they filter most of the stuff without ... I can switch frequencies and go back 50 years in behavior, ...
    (rec.travel.europe)