I just skimmed the article (very intersting graph showing the chronological improvement by JVC BTW).
Originally Posted by coldmachine
It certainly seems that many of us have used the term MTF in a reather nappropriate manner. I certainly use it to describe rapid transition, or geometry preservation, rather than a peak to peak value. The handling oftransition between states is far more apparent than an amplitude measure.
It seems clear that MTF, as used correctly, is simply unable to describe the sharpness, or inter pixel contrast, of digital units, and is even less able to numerically represent the visible difference between displays. I think we need a new metric to better describe this.
You know, I'm not so sure I agree, but maybe we'll just be arguing semantics.
The problem seems not to be that MTF is the wrong metric, measurement, the problem seems to be that we are unable to create or measure sinusoidal measurement at sufficient frequency to capture the "complete" MTF.
What I mean is we really need to run MTF out to a spacial frequency of somewhere seven or more times 1920 lines per picture width, or 960 line pairs per width. If we could run the measurements out to say 9600 line pairs per pixel width, I think we'd see the MTF graphs clearly show the "transition speed" difference between these projectors.
The problem is we're limited by the fact that the chips can only display point samples, and can only represent them as pseudo square wave. It's the fact that the (at least for the sake of discussion) "ideal" representation of the sample is a constant value across it's area, that I say we need 5 to 10x the frequency to see the difference. Square waves have significant harmonics out well past 5x, even toward 7x or 9x.
This is why I'm very curious about the Fourier Transform method of calculating MTF that Mark has touched on. It would seem, to my "haven't dug into it myself" eyes that since we know the ideal response is a square wave, and we can measure the actual response, we could do the transform and come up with the "frequency response"/MTF of the entire system out well past the "1080p" limit.
It seems it should be little different than what we do to capture room response with audio, we take a know input signal, in this case a 1080p alternating line pattern, which is a square wave of frequency 960 line pairs per picture width, and we know the result from Mark's measurements.
If I remember my communication systems class right, you convolve these two spacial domain signals which results in the transfer function of the system (which would ideally be an impulse, but it won't be because this is the real world). You then perform the Fourier Transform on the transfer function and you have the frequency domain response of the system.
Of course this is also prime to light of another thread
, but what is the theoretically ideal response? What does the information on a disc really represent? It it indeed a constant value of a finite area, ie a square, or is it the value of a point (no area). And please, if we want to discuss this, lets go to another thread so as not to derail this excellent fact-based thread.