Nyquist vs framerate

[miamicanes]

Anyone who's been involved with digital audio knows all about Nyquist and the rule that the sample rate needs to be at least twice the highest frequency to avoid artifacts. A commonly-accepted corollary is that the exact sample rate itself becomes irrelevant as long as each sample covers a period less than half the period of the highest-frequency. For audio, with sample rates > 44,000/sec, it seems to work fine.

On the other hand, anyone who's tried testing the theory with video knows that it seems to completely break down with regard to video fieldrate vs film framerate. According to Nyquist, as long as the video framerate is equal to or greater than 48 progressive frames per second, the exact framerate (or even the regularity of its periods) shouldn't matter. 24fps source shown at 60fps should be indistinguishable from 24fps source shown at 72fps. Of course, that's NOT the case at all.

So... at what point DOES precise framerate become increasingly irrelevant? at 4X (96fps and above) the framerate? Above some absolute framerate that has nothing to do with its factor of 24 and instead exceeds some threshold of human vision? And why DOES Nyquist seem to break down so badly for 24fps film? Is it because 24fps, and even 48fps, are both grossly below the framerate REALLY necessary for an indistinguishable illusion of motion? Or is it because 24fps film has a "true" framerate that's much, much higher than 24fps due to the passive encoding of additional visual information by things like motion blur, so the TRUE Nyquist threshold is actually twice the information rate conveyed by THAT?

My personal theory is that it's probably a combination of all the above... with motion blur effectively doubling the effective framerate of film to 48hz (necessitating 96hz or above for Nyquist to hold and permit arbitrary framerates not exactly equal to a whole-number multiple of 48), and human vision itself demanding something like 85-120hz to truly provide an illusion that's indistinguishable from live motion.

Anyone else have any observations or theories about the matter?

[FOO]

I think you are trying to apply an inappropriate theory to
video. It's a different problem and the Nyquist criterion
doe not apply.

[edsmith77]

To follow on with FOO:

Nyquist criteria :
Physical law, named after its finder Nyquist, that states that the digital representation of an analog signal will never contain any frequencies that are higher than half of the sampling frequency.
Frequencies above the so-called Nyquist frequency will be folded back into the audible range and cause usually undesired artifacts ; Aliasing. ( Philips )

[oddbod77]

@FOO: Actually Nyquist theorem would appear to be appropriate, just not as important because of the way human vision operates.

miamicanes, have you been reading this article by any chance? I noticed your post about the NTSC Red Dwarf DVD and couldn't help but see a common thread between the two posts.

[trossin]

From what I can remember about the sampling stuff that I learned 20 years ago is that if you can sample a signal at a tad faster than 2x the highest frequency of original signal you will be able to completely reconstruct it if you have an IDEAL filter.

If you don't have an ideal filter you get crap. I believe the reason you see problems is due to the filter. There is no such thing as an ideal filter but some are pretty darn close in the 1D domain of audio. When you talk about 2D (image) filtering is much harder and therefore farther from ideal.

[cford]

There are a few things going on in video that *mostly* make Nyquist not appropriate:
- The output format. With audio, the end-to-end system not only includes the sampling - the conversion from analog to digital - but also includes the re-conversion to analog. There is a reconstruction filter that essentially rebuilds the signal to (theoretically) its original, analog form. Video counts on your eyes to do the final D/A conversion, to be that reconstruction filter/interpolater. In order for audio to compare to this, you'd have to play back the samples as digital bursts and hope the ear works it out. The Shannon sampling theorem always includes the reconstruction end - it says that if you have the samples, then you CAN reconstruct, but it doesn't just end at the samples. Our brain works well acting as a D/A for visual, but I don't know if we can know HOW well.
- The concept of frequency. With audio, the signal is sound waves. But what is it in video? Is it light, or visible wavelengths? No, it's change or motion from one frame to the next. Frequency implies periodicity, which is usually not applicable to movement or action on the screen. If someone is moving across the screen, and you only take two frame samples of it, it will appear jerky, but your mind will fill in the blank and there will be no aliasing. If, however, they move to one side of the screen and partway back, and you only get two shots, you may miss that they changed direction, and that would be aliasing. The best example of this happening, and the reason I used the word "mostly" in the first paragraph, is the effect of wagon wheels seeming to spin backwards. This is indeed aliasing, and takes place because there is a cyclic, periodic signal in the motion of the wheels.
So I think Nyquist can be applied in specific situations, but you have to be very careful about what you define as your signal.
FWIW...
Carolyn

[lordsmurf]

Unless your making products for countless thousands of people, I wouldn't think too much into it.