Re: OT: Rock and Chips

On 2 Feb, 17:13, Kennedy McEwen <r...@xxxxxxxxxxxxxxxxxx> wrote:
In article
<b52a3b19-919e-43cb-9376-c54c61940...@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
"davidrobin...@xxxxxxxxxxxxxxxx" <davidrobin...@xxxxxxxxxxxxxxxx> writes



On 1 Feb, 19:27, Kennedy McEwen <r...@xxxxxxxxxxxxxxxxxx> wrote:
In article
<6d5099b6-7db0-4582-a366-dad9fd6a7...@xxxxxxxxxxxxxxxxxxxxxxxxxxx>,
"davidrobin...@xxxxxxxxxxxxxxxx" <davidrobin...@xxxxxxxxxxxxxxxx> writes

"Correct" sampling is band limiting, _followed_ by sampling with a
unit impulse - an infinitely short pulse with unit area.

Wrong!  You are assuming that the band limiting filter and the sampling
process *must* be independent when there is *no* requirement for them to
be so.  A 40kHz sampling system with a 20kHz input bandwidth is
completely indistinguishable from a 20kHz low pass filter and an
instantaneous sampling system.  Similarly, a time averaged sample
sequence is indistinguishable from a continuous running average filter
multiplied with an infinitely narrow sampling pulse sequence.

Where the pulse isn't infinitely short, and hence the resulting
sampling "instant" is non-instantaneous (i.e. finite), this changes
the frequency response of the sampled system.

Indeed it does, because the time averaging is a just another temporal
filter, which can be the major part of a band limiting filter.

The correct filter is a sinc pulse. It has this shape...

http://zone.ni.com/cms/images/devzone/ph/3e5f7a871017.gif

Your suggested "filtering", averaging linearly over the sample
instant, is a simple rectangle.

Without addressing every point of your post, since even the Internet
still has limited bandwidth(!), suffice it to say that your initial post
on this thread's archive regarding sampling reads: ' "Correct"
"sampling" requires _instantaneous_ sampling. '.  You have now accepted
that "correct" sampling does indeed require some time averaging.

No it doesn't - it requires instantaneous sampling of a band limited
signal. If the sampling isn't instantaneous, it requires correction to
make it "correct".

You also claimed that Nyquist sampling "did not really apply to video"
on 29th January and now appear to accept that it is, as Shannon stated,
perfectly applicable, indeed included in his work from very first
publication.

You may be determined to confuse...
1. the spatial dimensions of an image
2. the temporal dimension of the analogue representation of a video
signal, and
3. temporal as in "subsequent moments in time" cf frames
....but like it or not, they're different, involving different
compromises. btw, you can trivially Nyquist sample 2.

However your claim that I am simply "applying Nyquist & Shannon theory
blindly" is completely absurd.  I have made it clear from in both of my
previous posts in this thread that time averaged sampling is an
approximation to ideal sampling (posts of 30th Jan and 1st Feb) which
significantly reduces the artefacts that are created by shorter samples,
such as the 180deg or narrower shutter which, at any frame rate, is
further undersampling the motion.

I am well aware of the limitations of time averaged sampling and have
stated those limiations in both previous posts, however 100% frame time
average, together with time averaged display as a reconstruction filter,
is far, far closer to ideal than the infinitely short sampling at the
same frame rate that you were claiming was "correct".

Now you're straying into fantasy - the idea that, in practice, time
averaged display is better than instantaneous display wrt motion
portrayal at video frame rates is demonstrably wrong.

 Furthermore, if
you examine the characteristics required for ideal sampling, ie.
temporal band limited (*not* necessarily flat, as per the sinc)

....if the frequency response isn't flat, it's not ideal.

you will
find that the 100% time average shutter is actually the closest that it
is practical to implement in both film and analogue video systems

This ^ is almost true, but...

: the
rect function being the fundamental of the sinc function.

....the closest rect function to the correct sinc function isn't 100%.
I have nothing to check it on, but I bet it's about 66%.

 As such, it
does exactly what Rod claimed for it - provides a greater realism with
vastly reduced motion artefacts over narrow angle shutters and
equivalent which are simply excessively undersampled in the time domain.

Maybe - most film makers don't seem to agree.

It's still not perceptually transparent Nyquist sampled. It's a bodge.

And ideal sampling requires band limiting and instantaneous sample
points.

Still, if you didn't believe me the first time I said it, why should
you now?

http://en.wikipedia.org/wiki/Sampling_function

Cheers,
David.



--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers         (replace 'nospam' with 'kennedym' when replying)

.

Relevant Pages

  • Re: OT: Rock and Chips
    ... that "correct" sampling does indeed require some time averaging. ... portrayal at video frame rates is demonstrably wrong. ... set a spoked wheel spinning and video it slowing down at different shutter speeds - the wagon wheel effect reduces with slower shutter speeds, ...
    (uk.tech.digital-tv)
  • Re: OT: Rock and Chips
    ...  You are assuming that the band limiting filter and the sampling ... You have now accepted that "correct" sampling does indeed require some time averaging. ... I have made it clear from in both of my previous posts in this thread that time averaged sampling is an approximation to ideal sampling which significantly reduces the artefacts that are created by shorter samples, such as the 180deg or narrower shutter which, at any frame rate, is further undersampling the motion. ...
    (uk.tech.digital-tv)
  • Re: Simple & small MCU with 4 UARTS
    ... software bit banging. ... What do you think is the upper baud limit for 1 to 4 software bit ... drops at the start of the start bit, your ideal sampling point is then ...
    (comp.arch.embedded)
  • Re: Simple & small MCU with 4 UARTS
    ... UART on a fast AVR before a hardware UART is needed? ... The key synchronisation point is the start bit - from when the line drops at the start of the start bit, your ideal sampling point is then half a baud time later. ...
    (comp.arch.embedded)