12 Apr
2004
12 Apr
'04
9:28 a.m.
Martin McCormick WB5AGZ Stillwater, OK
OSU Information Technology Division Network Operations Group
Subject: Re: [linux-audio-user] Fwd: [Jamin] Re: soft clip: Achieving Gain,
inconsequential overloads
--------
Steve Harris writes:
Actually thats not quite right - AD converters
generally work in 1's or
2's complement, depending on the brand/variety, so the middle point is at
0, and the range is slightly larger in the -ve voltage range.
Correct. We may be talking about the difference between
signed and unsigned number representation. If you have an unsigned
value, the only possibilities are from 0 to the maximum value
represented by all ones. There are no negative numbers. When you
have a signed value, then an 8-bit number, for example can be
represented by +1 to +127 or -1 to -128.
Now, here is something I am a bit unsure of, myself. I know,
from actual observations that a straight PCM output from your basic
A/D converter if read as unsigned numbers moves in steps from 0 to all
1's on. I honestly have not tried to interpret those data as signed
numbers because it wasn't convenient at the time. If one wants to
have a valid representation of what the wave form is doing for
graphical or calculation purposes, then the mid-point would have to be
what one would call 0 level with -1 being one below and 1 being 1
above, etc. I am certainly not arguing with anyone, but am a wee bit
confused as to the correct way to represent the numbers.
This is probably off-topic, but any graphical software that shows
you your music wave forms or does DSP functions in an arithmetic
manner that is based upon an AC model has to behave as if the
mid-point value was 0.
Maybe someone can set me/us all straight in this manner.
G
12 Apr
12 Apr
2:34 p.m.
On Mon, Apr 12, 2004 at 06:21:33AM -0500, Martin McCormick wrote:
Now, here is something I am a bit unsure of, myself.
I know,
from actual observations that a straight PCM output from your basic
A/D converter if read as unsigned numbers moves in steps from 0 to all
1's on. I honestly have not tried to interpret those data as signed
numbers because it wasn't convenient at the time. If one wants to
have a valid representation of what the wave form is doing for
graphical or calculation purposes, then the mid-point would have to be
what one would call 0 level with -1 being one below and 1 being 1
above, etc. I am certainly not arguing with anyone, but am a wee bit
confused as to the correct way to represent the numbers.
Audio AD converters are asymmetrical - abs(minimum_voltage) is higher than
maximum_voltage. Audio AD converters sometimes have unsigned modes, but
they seem to have different behaviour. I checked up recently to make sure
JACK was doing the right thing(TM) when converting between floats and
ints. This stuff is explained in detail in the AD converter spec sheets
for the AD converters uned in the m-audio delta cards.
This is probably off-topic, but any graphical
software that shows
you your music wave forms or does DSP functions in an arithmetic
manner that is based upon an AC model has to behave as if the
mid-point value was 0.
The midpoint value in 2's complement is -0.5bits, which does not represent
0volts.
- Steve
2:59 p.m.
On Mon, 2004-04-12 at 09:27, Steve Harris wrote:
On Mon, Apr 12, 2004 at 06:21:33AM -0500, Martin
McCormick wrote:
Steve,
I wonder if you have any info on something I was reading in Bob
Katz's book recently, that many A/D converters actually have compressors
or other sorts of non-linear circuits built into them which effect the
use of the converter when it gets anywhere near maximum value anyway.
I found some of his comments about what happens with A/D's in these
regions interesting. Do the spec sheets you've looked at imply anything
like this?
Thanks,
Mark
Now, here is something I am a bit unsure of,
myself. I know,
from actual observations that a straight PCM output from your basic
A/D converter if read as unsigned numbers moves in steps from 0 to all
1's on. I honestly have not tried to interpret those data as signed
numbers because it wasn't convenient at the time. If one wants to
have a valid representation of what the wave form is doing for
graphical or calculation purposes, then the mid-point would have to be
what one would call 0 level with -1 being one below and 1 being 1
above, etc. I am certainly not arguing with anyone, but am a wee bit
confused as to the correct way to represent the numbers.
Audio AD converters are asymmetrical - abs(minimum_voltage) is higher than
maximum_voltage. Audio AD converters sometimes have unsigned modes, but
they seem to have different behaviour. I checked up recently to make sure
JACK was doing the right thing(TM) when converting between floats and
ints. This stuff is explained in detail in the AD converter spec sheets
for the AD converters uned in the m-audio delta cards.
3:20 p.m.
On Mon, Apr 12, 2004 at 09:53:06AM -0700, Mark Knecht wrote:
Steve,
I wonder if you have any info on something I was reading in Bob
Katz's book recently, that many A/D converters actually have compressors
or other sorts of non-linear circuits built into them which effect the
use of the converter when it gets anywhere near maximum value anyway.
I found some of his comments about what happens with A/D's in these
regions interesting. Do the spec sheets you've looked at imply anything
like this?
They might have but I dont remeber, you can find the spec sheets by
googling for the part numbers.
Thats interesting information. I wonder how consistent the compression is.
- Steve
4:06 p.m.
On Mon, 2004-04-12 at 10:14, Steve Harris wrote:
On Mon, Apr 12, 2004 at 09:53:06AM -0700, Mark Knecht
wrote:
My guess would be not at all consistent. However, I think it's germane
to Ron Parker's thread about getting things louder. Katz, in his book,
pushes you to see this as one problem in getting to the next level
sonically. It's not only a problem in the A/D, but apparently also in
some mix buses inside of DAWs too.
As I've sort of soaked up all the information he has to offer in that
book, (and it's a lot,) it's led me to understand that there are a lot
of reasons to potentially keep the tracking volume pretty far down, but
it does lead to more problems when people want loud on the final CD.
As an example of how good things can get when you don't make a CD loud,
check out MyCoy Tyney - New York Reunion. Katz won a Grammy for this
recording. Granted, I have to turn my stereo up, but sonically it's one
of the best I've found in a long, long time. (Available as a hybrid for
you SACD listeners out there.)
Steve,
I wonder if you have any info on something I was reading in Bob
Katz's book recently, that many A/D converters actually have compressors
or other sorts of non-linear circuits built into them which effect the
use of the converter when it gets anywhere near maximum value anyway.
I found some of his comments about what happens with A/D's in these
regions interesting. Do the spec sheets you've looked at imply anything
like this?
They might have but I dont remeber, you can find the spec sheets by
googling for the part numbers.
Thats interesting information. I wonder how consistent the compression is.
5:07 p.m.
On Mon, Apr 12, 2004 at 09:53:06AM -0700, Mark Knecht wrote:
I wonder if you have any info on something I was
reading in Bob
Katz's book recently, that many A/D converters actually have compressors
or other sorts of non-linear circuits built into them which effect the
use of the converter when it gets anywhere near maximum value anyway.
It's *extremely* desirable for any clipping do be done in the analogue
domain before it reaches the A-D converter.
Clipping produces rich out-of-band harmonic content which gets aliased
to in-band content if the clipping is done by the A-D converter
itself. The analogue signal should get clipped before it enters the
anti-aliasing fiter, so those harmonics are removed. Any in-band
harmonics remain, of course.
This may explain why overloading an A-D can sound so bad - it's worse
than razor sharp analogue signal clipping.
Incidentally, the analogue clipping level should be a little below
should be below full scale, to allow for the anti-aliasing filter to
overshoot. Ideally the maximum overshoot level should remain within the
range of the A-D.
I have no idea how many of the commercially available audio A-D
converters do this properly. I suspect that low-cost sound cards don't
even have an anti-aliasing filter at all.
Adding a compressor or fast-acting limiter is another matter. Again I
don't know if any do this. I would have thought it was compromising the
signal quality, but you could argue that the user is asking for trouble
if he sets the levels too high anyway, and that the limiter has no
effect on lower level signals.
--
Anahata
anahata(a)treewind.co.uk -+- http://www.treewind.co.uk
Home: 01638 720444 Mob: 07976 263827
5:26 p.m.
On Mon, 2004-04-12 at 11:27, Anahata wrote:
On Mon, Apr 12, 2004 at 09:53:06AM -0700, Mark Knecht
wrote:
Certainly. Any clipping is bad, whether digital or analog, but given the
choice I'd far rather do it in the analog domain.
One problem that I'm faced with on the Pro Tools 002 R is that it has a
nice preamp, and a nice converter, but no way to patch a compressor in
the middle. For this reason I decided to record at pretty low levels to
give me enough headroom to not clip when the drummer laid into his snare
or kick.
You could correctly argue that recording at low volumes reduces your
SNR, which is true, but unless you're recording in a professionally
built studio my experience is that normal room noise from AC, fans,
lights, etc. is way above the noise floor anyway.
<SNIP>
I wonder if you have any info on something I
was reading in Bob
Katz's book recently, that many A/D converters actually have compressors
or other sorts of non-linear circuits built into them which effect the
use of the converter when it gets anywhere near maximum value anyway.
It's *extremely* desirable for any clipping do be done in the analogue
domain before it reaches the A-D converter.
Adding a compressor or fast-acting limiter is another matter. Again I
don't know if any do this. I would have thought it was compromising the
signal quality, but you could argue that the user is asking for trouble
if he sets the levels too high anyway, and that the limiter has no
effect on lower level signals.
One of Katz's points, and I'm sure this is based on his experience and
on specific A/D's, is that a full level input signal may actually be
coming close to the power supply level that the A/D is operating at.
I.e. - if I'm accepting +/-5v inputs to get full scale, but the chip is
operating on a 5V power supply, then the converter circuits will most
certainly go non-linear. However, if full scale input is +/-5 volts but
the A/D is operating on a 12V supply then the circuits can remain linear
even at full scale.
- Mark
7559
days inactive
7559
days old
linux-audio-user@lists.linuxaudio.org
6 comments
4 participants
participants (4)
-
Anahata -
Mark Knecht -
Martin McCormick -
Steve Harris