Re: low-pass filter question
Re: low-pass filter question
- Subject: Re: low-pass filter question
- From: Herbie Robinson <email@hidden>
- Date: Tue, 17 Jan 2006 19:35:45 -0500
To throw this out there, are there any plans for
the CA team to ever enhance or improve the SRC?
I was always personally completely satisfied
with what it is, and honestly I'd take the
convenience of the CA SRC (in terms of coding)
over the possible ±20dB of clarity found at
<-120dB for other methods. I've never yet had a
user complain about the SRC quality...
...or is there really no need when it comes to
the true floating-point nature of the CA SRC? Is
it possible that the "noise" in the charts comes
just as much from the 24-32-24 bit conversion?
Well, let's look at the math involved. The SRC
is basically a long FIR filter. Let's say 100
taps for ease of computation. The computation
for each sample goes like this:
Conversion to floating point.
FIR Filter
Conversion to fixed point.
The conversion of a 24 bit integer sample to
floating point is lossless (even in single
precision).
The FIR filter consists of 100 multiplies and 100
adds. Each one of these will accumulate a
roundoff error. Using single precision SSE2
instructions, all of those roundoff errors are
roughly at the 24 bit level and the roundoff
errors add up -- That corresponds to distortion
being in the low order 7 bits of the floating
point mantissa. This can be improved upon a lot
by sorting the coefficients; so, it's not as bad
as it sounds, but there still has to be quite a
bit of distortion build-up.
The conversion from float to fixed introduces one
more roundoff error. If the conversion is
dithered (it should be), then there will also be
some white noise added at this stage. Note that
ALL the SRCs in the tests are doing this
conversion; so, the threshold in the DSP must be
set higher than this level of noise (or none of
the tests be showing those large areas of blue).
I may or may not be an audio purist, but at
levels as low as -120dB or less, have any tests
shown that that stuff is actually *audible* in
the end result? Is this just grandstanding in
terms of numbers and charts, or does this have
real audible validity?
Surprisingly enough, artifacts in the low order
24th bit are audible, but only barely. I don't
know of any formal studies, because it's
difficult (i.e., expensive) to test
psycho-acoustic differences at this level and the
studies that get funding are along the lines of
"what will some percentile 12 year old using $5
ear buds tolerate without complaining?" rather
than "what can people with really good hearing
detect?" or "how does the long term exposure to
various kinds of truncation error affect the
perception of musical content".
The only experiment I have done involved
comparing two mixes I prepared myself. One mix
had dither applied at the 24th bit level by the
mixer and one did not. I found that I could
identify a difference a significant amount of
times in a blind test. I had no preference for
one sounding better than the other, but I could
just barely hear a difference. The material in
question was recorded in a very nice acoustic
setting and was mixed with no processing other
than the mixer, using nothing but natural
ambience and I was listening on electrostatic
headphones. The methodology was ABX, with very
short times between samples (to make sure the
acoustic memories didn't fade). I was doing it
myself; so, it couldn't be totally blind, but I
put together a DAW session with the test in it
and waited a day so I had hopefully forgotten
what was in it.
This is a difference in only the lowest bit of a
24 bit samples and my hearing is anything but
pristine (30+ years of clubbing and playing in
rock and blues bands). It's been pretty easy for
me to hear the difference between 16 and 24 bit
material now that I know what the difference is,
but it's very hard to describe it. The only
thing that seems to work is visual analogies,
like "veiled". Also, it seems to be more
relaxing to listen to 24 bit material.
The general conjecture about this sort of thing
is that the noise (or distortion or whatever you
want to call it) is very un-natural. It isn't
random (like white noise) and it isn't harmonic
distortion. It isn't even like the Bessel
function related partial series that come off
drum heads. It looks like the modulation of the
original signal by the sampling frequency and
that aliases back into the audible range. The
result is signal correlated non-harmonic content
that our ears aren't used to processing. Some
people are really irritated by it, but most
people can't hear it until they get to compare 16
and 24 bit samples.
--
-*****************************************
** http://www.curbside-recording.com/ **
******************************************
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Coreaudio-api mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden