skotdee, let me try to explain what I meant.
Almost everyone is familiar with the VU meter--it has a scale with levels marked from -20 up through 0 dB, and then you can go 3 dB above which is "in the red." That corresponds to classic analog transmission and recording media, where 0 VU wasn't a brick-wall limit. In systems of that kind, to maximize your signal-to-noise ratio, occasionally you could (and even should) let the extreme peak signals go briefly "into the red"--but the needle was never supposed to hover or stay there, or hit the "pin" (mechanical stop at the extreme), or else the distortion might very well become audible.
Digital audio levels operate on a fundamentally different principle. For example if I ask you to tell me the highest two-digit number in decimal, clearly that is 99, and there's no possible way of bargaining with that fact. Add one more and you've definitely exceeded the range of two-digit decimal integers.
Digital audio levels are like that--they have a set number of bits allocated for the sample values, and if you go one beyond the limit, the system no longer has any valid way of recording that value in the number of bits that it has, so it has already begun to distort. The farther you try to exceed the limit, the more the distortion becomes. There's no "red zone" in the sense of a transitional zone in which brief peaks are OK, in other words. And in a strict sense there would also be no such thing as digital overload, since the highest possible sample value simply is whatever it is; no sample value greater than that can possibly be present in the bitstream because there's no way to write it in the number of bits you have (just as you can't write 100 with only two decimal digits).
So, when your software is showing "overloads" it can only be inferring them on the basis of a succession of some number of samples that are not in themselves overloads, but which are at the maximum value for the available number of bits. For many professional digital audio equipment the usual threshold for indicating an overload has been three or more consecutive full-scale values. (Actually the Red Book CD audio standard imposes an additional limit, but I'll leave that out unless someone specifically wants to know about it.)
Now, if I recall correctly what the heck we're talking about, someone referred to a graphical reconstruction of a waveform from digital samples, showing overloads between the samples. That is entirely possible, but all such displays must be based on certain assumptions about the way in which the signal will be recovered (converted from digital to analog) because to put it mildly, not all D/A converters handle such signals the same way. One big issue is that the digital filters used in modern D/A converters have some digital gain which is not necessarily unity; as a result, many of them can even overload occasionally even if NO sample values reach digital full scale. Each such filter has its own "worst case" (nightmare scenario) succession of sample values which may push it as much as 5 - 6 dB over the edge.
The analog circuitry attached to the digital circuitry is another matter; sometimes there's not enough headroom there. But for now I'm speaking strictly of the digital hardware in the particular converter, and how it responds to perfectly valid (though statistically very unlikely) trains of sample values. That's why I said that this all depends on the D/A that's chosen, either for actual analog output or as the model which a given graphical display in software will emulate.
--best regards
Topic: clipping question (Read 2409 times)