[Scott Dorsey]

Frank Stearns wrote:

Don't you need some sort of sample value delta, even at 1 or 2 Hz, to have a HPF do
something? Doesn't a constant bias, such as DC, merely shift all sample values? And
if nothing's "moving", even at VLF, there's nothing for a digital HPF to find?

Think about the digital IIR filter as just being like an analogue high pass,
like a blocking capacitor and a shunt resistor. If you apply a DC offset to
the input with no signal, the filter will eventually reduce it down to zero
in some time that depends on the time constant of the filter. So you may have
a couple seconds of time before the offset is actually removed.

Arithmetically speaking, how does a digital HPF recognize "non-moving" DC offset and
not, say, do something inappropriate to a naturally very asymmetrical waveform, such
as a brass instrument? Often brass instrument waveforms look like they have DC in
them, when in fact that's what brass beasties produce in real life.

The same way that an analogue HPF filter does... and that indeed can cause the
DC point to be shifted while the insrument is playing, but it doesn't
necessarily make anything more symmetric.

Why does the DC offset removal tool in Sound Forge, for example, require at minimum
5 seconds worth of samples (or a scan of the entire sample set of the file) to do DC
removal?

Probably because the filter has some settling time. But also, the DC offset
removal tool may be something other than a simple high pass as well. That
doesn't mean that a simple high pass isn't a reasonable solution though.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."