[opendtv] Re: Mediaflo Technology
- From: Craig Birkmaier <craig@xxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Sat, 11 Nov 2006 08:25:28 -0400
Looks like Tom caught you here Mark.
You can't compare apples with oranges. We are talking about Nyquist
limited images in both cases - i.e. video.
As we have discussed MANY TIMES in the past, the energy in these
images is spread out over many pixels. For that "peaked" black line
we are talking about at least three, and probably five or more. The
peak will move through the digital grid just as it moves through the
shadow mask of the analog TV. You are correct that the analog signal
allows for improved edge positioning, but modern codecs are moving to
quarter pel positioning in the encoder to deal with these issues as
well. Sadly, we rarely watch analog TV on a B&W monitor, where we can
take real advantage of the detail possible with an analog CRT.
By the way, a "digital display" has the ADVANTAGE in that it can also
display non-Nyquist limited images for locally generated graphics.
Even more important, you have left out the all important optical
filter in the acquisition device that would capture the image. This
filter will assure that the energy of that edge will be spread out to
allow that peak detail to move through the sample grid without
aliasing.
Bottom line, it's all about filtering and the benefits of
oversampling ahead of the encoder.
Regards
Craig
And one more thing - protecting the integrity of samples. MPEG
quantization robs detail from an image, just like distortion and
noise degrade an analog transmission.
At 4:41 PM -0500 11/10/06, Tom Barry wrote:
Mark -
I think in order for nyquist to apply properly (such that you can
calc with it) you would have to have the lines fading back to white
and back in a sin wave, not with sharp edges which already imply
higher frequencies.
And the Nyquist limit for that for, say, at 720 pixels would have to
be LESS than 720, say 718 lines or 359 complete cycles. But I
believe a Fourier transform of these 359 (sinusoid) line pairs using
720 samples would indeed capture all the information, regardless of
phase. So you would not have the blinking effect as something
scrolled very slowly, or the confusion of motion compensation due to
aliasing. And if you upscaled the subsequent sampled (but nyquist
limited) image you could still get an exact representation of the
original. But, again, this assumes the magnitude of the signal AT
the Nyquist frequency and all higher frequencies was originally zero.
Unfortunately I'm not sure the convenient discrete cosine transform
used in the various MPEG's has this property. And turning things
into a 8x8 block based transform further limits the resolution to
7/8 of the nyquist frequency, much worse than my 359/360 in my
example above.
Of course all this is only an approximation since all the sampling
theory assumes we are talking about a periodic waveform, which in
turn probably implies the real original image (at much larger
resolution) really contains only integer multiples of the sampling
frequency. And that isn't generally true in any real image anyway.
This means I'm still not so sure about the difference between
digital and analog in this regard. It's hard to get a grip on some
of this stuff.
- Tom
Mark Schubin wrote:
Consider a pair of vertical lines, one black and one white, each
the same width as a digital pixel. An ideal digital system would
just barely be able to reproduce them (they'd be at the Nyquist
limit). An analog system of the same Nyquist bandwidth would also
reproduce them, though sinusoidally instead of as black and white
stripes.
Now shift the lines one-half pixel horizontally. The analog system
has no trouble moving the sine wave by a 90-degree phase shift.
The digital system, however, is now all gray. Each pixel is
getting both a black value and a white value.
Thus, Ron's 440 pixels per line actually compares favorably to a
digital system with 880 (which is more than Rec. 601's 720).
TTFN,
Mark
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word
unsubscribe in the subject line.
Other related posts:
