3 Nov
2002
3 Nov
'02
12:13 p.m.
found out some interesting facts about the chebyshev. been playing
around a little with a chebyshev shaper, feeding it various harmonic
amplitudes and a sine oscillation, taking an FT afterward.
it seems that in order to get a harmonic of amplitude 0.5, you must
not pass 0.5 to chebpc for that harmonic, but sqrt(2)/2. for a
harmonic of amplitude .333333333, pass in sqrt(3)/3, etc. thus, if we
want amplitude 'a' for a given harmonic, the chebpc coefficient is 'a
* sqrt (1/a)' rather than simply 'a'. (negative amplitudes: ?).
the peak value of the chebyshev-shaped output will be the sum of all
coefficients calculated in this manner. the further the incoming sine
is scaled down (from [-1,+1]), the less the harmonic mix will match
the wanted amplitudes.
for the amp code, this would mean we should probably try the
following: compress/expand the incoming signal to fit exactly into
[-1, 1] (normalize). the coefficient tables need some treatment, too
-- we want the output sum to be 1, and the relative strengths of the
harmonics to match.
i'm not up to understanding all implications of the fact that the
incoming signal is not a pure sine; neither do i have a recipe for
preparing the coefficient tables -- if we scale the individual
coefficients by 1/sum their mix will not match what we want.
tim
3 Nov
3 Nov
12:38 p.m.
On Sun, Nov 03, 2002 at 05:08:33 +0100, Tim Goetze wrote:
the peak value of the chebyshev-shaped output will be
the sum of all
coefficients calculated in this manner. the further the incoming sine
is scaled down (from [-1,+1]), the less the harmonic mix will match
the wanted amplitudes.
Hmm, I suspect that this sames the cheby unsiutable for what we want, the
output from a guitar appears to be far fronm a sin.
for the amp code, this would mean we should probably
try the
following: compress/expand the incoming signal to fit exactly into
[-1, 1] (normalize). the coefficient tables need some treatment, too
-- we want the output sum to be 1, and the relative strengths of the
harmonics to match.
i'm not up to understanding all implications of the fact that the
incoming signal is not a pure sine; neither do i have a recipe for
preparing the coefficient tables -- if we scale the individual
coefficients by 1/sum their mix will not match what we want.
I think we should probably abandon the cheby approach for now, this
combined with the freqency dependency problem probably makes it a looser
:(
- Steve
3:11 p.m.
Steve Harris wrote:
yep, it doesn't seem to be getting easier the more we look at it.
the above approach could bring us close, but it does need xfading
and two chebyshev shapers.
been thinking about how to do a hard clipper with sinc some more
today, without real results though.
tim
On Sun, Nov 03, 2002 at 05:08:33 +0100, Tim Goetze
wrote:
if we normalize the incoming signal, apply the shaper(s), then scale
down, it just might work -- it will get the harmonic mix right at
least.
the peak value of the chebyshev-shaped output
will be the sum of all
coefficients calculated in this manner. the further the incoming sine
is scaled down (from [-1,+1]), the less the harmonic mix will match
the wanted amplitudes.
Hmm, I suspect that this sames the cheby unsiutable for what we want, the
output from a guitar appears to be far fronm a sin. i'm not up
to understanding all implications of the fact that the
incoming signal is not a pure sine; neither do i have a recipe for
preparing the coefficient tables -- if we scale the individual
coefficients by 1/sum their mix will not match what we want.
I think we should probably abandon the cheby approach for now, this
combined with the freqency dependency problem probably makes it a looser
:(
7:24 p.m.
On Sun, Nov 03, 2002 at 08:06:53 +0100, Tim Goetze wrote:
yep, it doesn't seem to be getting easier the more we look at it.
the above approach could bring us close, but it does need xfading
and two chebyshev shapers.
We'd also need to split up the incoming signal by frequency though, and
that would make it expensive to run.
i'm not up to understanding all implications of
the fact that the
incoming signal is not a pure sine; neither do i have a recipe for
preparing the coefficient tables -- if we scale the individual
coefficients by 1/sum their mix will not match what we want.
I think we should probably abandon the cheby approach for now, this
combined with the freqency dependency problem probably makes it a looser
:( been thinking about how to do a hard clipper with sinc
some more
today, without real results though.
Yeah, I think thats difficult, and probably not neccesary, the "hard clip"
from a guitar amp doesn;t look very hard to me, so I recon you could just
apply the shaper, plus a bit of oversampling, a LP filter and it'd be
fine.
Remind me: what was the thing that put us off chains of models? Was it
just the messy bottom end? Or was there more? I'm pretty sure I can fix
the muddly low harmonics and it sounds like other people aren't modeling
the variation in harmonic amplitude with signal amplitude, so maybe it
isn't important.
- Steve
4 Nov
4 Nov
11:45 a.m.
Steve Harris wrote:
http://quitte.de/driven.gif
does look like hard clipping to me (sine, shaped at maximum gain).
We'd also need to split up the incoming signal by
frequency though, and
that would make it expensive to run.
very much so.
been thinking
about how to do a hard clipper with sinc some more
today, without real results though.
Yeah, I think thats difficult, and probably not neccesary, the "hard clip"
from a guitar amp doesn;t look very hard to me, so I recon you could just
apply the shaper, plus a bit of oversampling, a LP filter and it'd be
fine. 
Remind me: what was the thing that put us off chains of
models? Was it
* the harmonic mix didn't resemble the original
* too muddy
* aliasing when stacked (though oversampling might help)
just the messy bottom end? Or was there more? I'm
pretty sure I can fix
the muddly low harmonics and it sounds like other people aren't modeling
the variation in harmonic amplitude with signal amplitude, so maybe it
isn't important.
went looking at some pod distortion. it seems you're right about
the harmonic mix not being influenced by signal amplitude, it
seems they just fade them in with more gain. an interesting fact is
that the harmonics it generates die away before they reach 11 kHz; a
high f sine input will cause relatively few harmonics, a low f will
have many. looks like they simply use a LP (albeit one with very good
stop-band attenuation) to fight aliasing.
i'm puzzled how they do manage to come up with such a lot of
different distortion characters though.
tim
12:10 p.m.
On Mon, Nov 04, 2002 at 04:24:19 +0100, Tim Goetze wrote:
Yeah, that'l be allright <waves hand nonchalanantly> ;) if we were talking
square waves I'd be worried.
Yeah, I think
thats difficult, and probably not neccesary, the "hard clip"
from a guitar amp doesn;t look very hard to me, so I recon you could just
apply the shaper, plus a bit of oversampling, a LP filter and it'd be
fine.
http://quitte.de/driven.gif
does look like hard clipping to me (sine, shaped at maximum gain). went looking at some pod distortion. it seems
you're right about
the harmonic mix not being influenced by signal amplitude, it
seems they just fade them in with more gain. an interesting fact is
that the harmonics it generates die away before they reach 11 kHz; a
Yeah, that hints very strongly at cheap antialising (like we're planning)
- I think its fine, guitar cabinets and cones aren't that big on high
frequencies anyway.
i'm puzzled how they do manage to come up with
such a lot of
different distortion characters though.
My current guess is different transfer functions. Do you know the phrase
"when all you have is a hammer, every problem looks like a nail" ;)
- Steve
2:06 p.m.
On Mon, Nov 04, 2002 at 04:06:53PM +0000, Steve Harris wrote:
My current guess is different transfer functions. Do
you know the phrase
"when all you have is a hammer, every problem looks like a nail" ;)
I'd bet money that they use different pre- and post- filters, too.
It's very instructive to play with this setup of effects pedals:
graphic EQ -> distortion -> graphic EQ
For instance, giving a heavy pre-boost to the lows / low mids
will make the sound "fuzzier" but soften the attack.
Think 1970s metal... sabbath, dude!
The resulting mud can be somewhat reduced by lowering the
bass on the "post" EQ.
Conversely, cutting the pre bass, and boosting the post bass,
will result in a sound that still has a lot of low end
but has a much sharper attack.
--
Paul Winkler
http://www.slinkp.com
"Welcome to Muppet Labs, where the future is made - today!"
2:43 p.m.
On Mon, Nov 04, 2002 at 10:03:11 -0800, Paul Winkler wrote:
On Mon, Nov 04, 2002 at 04:06:53PM +0000, Steve Harris
wrote:
Absolutly. I also remeber discussing speaker modelling with a hardware guy
and he said that EQ was the most important part of that. Tim's bigass IIR
hinted at that too.
My current guess is different transfer functions.
Do you know the phrase
"when all you have is a hammer, every problem looks like a nail" ;)
I'd bet money that they use different pre- and post- filters, too.
It's very instructive to play with this setup of effects pedals:
graphic EQ -> distortion -> graphic EQ For instance, giving a heavy pre-boost to the lows /
low mids
will make the sound "fuzzier" but soften the attack.
Think 1970s metal... sabbath, dude!
...
Interesting, thanks.
Do you have any thoughts on trying to gather transfer functions with saws by
any chance? I'm still vaguely concerned about phase issues, but I guess we
wont know till we try it.
For the record my current guess for an amp process is:
.-> bandpass -> shaper -. .--- LP <--.
| | v |
input -> EQ? -+-> bandpass -> shaper -+-> delay -> EQ ---> output
| |
`-> bandpass -> shaper -'
With N bandpass filters, maybe 2 octaves each. The top and bottom could be
shelves, but I think you get DC block and antialiasing for free if you
dont bother. The EQ? could be optional if you have enough bandpasses.
I think it would be pretty fast. I'm hoping there is a proffesional amp
modeller out there who will eventually crack and admit how its done ;)
- Steve
3:13 p.m.
On Mon, Nov 04, 2002 at 06:39:33PM +0000, Steve Harris wrote:
Do you have any thoughts on trying to gather transfer
functions
with saws by any chance?
Circular saw? Skillsaw? Seriously, this stuff is beyond me.
I know diddley-squat about DSP.
For the record my current guess for an amp process
is:
.-> bandpass -> shaper -. .--- LP <--.
| | v |
input -> EQ? -+-> bandpass -> shaper -+-> delay -> EQ ---> output
| |
`-> bandpass -> shaper -'
well, we'd have to try it and see :)
the delay w/ lp in feedback - is that basically
how you make a resonant lopass filter?
dont bother. The EQ? could be optional if you have
enough bandpasses.
Sure, if there is pre-shaper and post-shaper gain for each
band then that's your EQ. Lot of parameters to control,
though.
--
Paul Winkler
http://www.slinkp.com
"Welcome to Muppet Labs, where the future is made - today!"
6:35 p.m.
On Mon, Nov 04, 2002 at 11:10:14 -0800, Paul Winkler wrote:
No, its a primitive cabinet simulator. Just model reflections off the back
wall. The LP represents absorbion in the fiberous stuff inside. I read
about the technique somewhere... music-dsp probably.
Well, they would be preset for given amp types I guess.
- Steve
For the record
my current guess for an amp process is:
.-> bandpass -> shaper -. .--- LP <--.
| | v |
input -> EQ? -+-> bandpass -> shaper -+-> delay -> EQ ---> output
| |
`-> bandpass -> shaper -'
well, we'd have to try it and see :)
the delay w/ lp in feedback - is that basically
how you make a resonant lopass filter? dont bother.
The EQ? could be optional if you have enough bandpasses.
Sure, if there is pre-shaper and post-shaper gain for each
band then that's your EQ. Lot of parameters to control,
though. 
6:45 p.m.
Make a pretty picture like that and make the shapers compressors and then I
have a Waves C4, the best little end-of-the-line compressor around. ;-)
-----Original Message-----
From: linux-audio-dev-admin(a)music.columbia.edu
[mailto:linux-audio-dev-admin@music.columbia.edu]On Behalf Of Steve
Harris
Sent: Monday, November 04, 2002 2:32 PM
To: linux-audio-dev(a)music.columbia.edu
Cc: Paul Winkler
Subject: Re: [linux-audio-dev] Re: Cheby amp code
On Mon, Nov 04, 2002 at 11:10:14 -0800, Paul Winkler wrote:
No, its a primitive cabinet simulator. Just model reflections off the back
wall. The LP represents absorbion in the fiberous stuff inside. I read
about the technique somewhere... music-dsp probably.
Well, they would be preset for given amp types I guess.
- Steve
For the record
my current guess for an amp process is:
.-> bandpass -> shaper -. .--- LP <--.
| | v |
input -> EQ? -+-> bandpass -> shaper -+-> delay -> EQ ---> output
| |
`-> bandpass -> shaper -'
well, we'd have to try it and see :)
the delay w/ lp in feedback - is that basically
how you make a resonant lopass filter? dont bother.
The EQ? could be optional if you have enough bandpasses.
Sure, if there is pre-shaper and post-shaper gain for each
band then that's your EQ. Lot of parameters to control,
though.
7:43 p.m.
On Mon, Nov 04, 2002 at 02:36:07 -0800, Mark Knecht wrote:
Make a pretty picture like that and make the shapers
compressors and then I
have a Waves C4, the best little end-of-the-line compressor around. ;-)
Funnily enough that was what made me draw it that way, I've been thinking
about multiband comressors in odd moments. I want some more feedback on
the quality of my compressor core before I embed it in anything that fidly
though.
- Steve
6:57 p.m.
On Mon, Nov 04, 2002 at 10:31:37PM +0000, Steve Harris wrote:
Of course. But then there will be things you want to
adjust - typical amp front-panel stuff:
pre, low, mid, hi, post at least...
the tricky bit would be mapping e.g. .the midrange control
onto simultaneous adjustment of various these "hidden" paramenters.
--
Paul Winkler
http://www.slinkp.com
"Welcome to Muppet Labs, where the future is made - today!"
Sure, if there
is pre-shaper and post-shaper gain for each
band then that's your EQ. Lot of parameters to control,
though.
Well, they would be preset for given amp types I guess.
9:04 p.m.
Steve Harris wrote:
On Mon, Nov 04, 2002 at 11:10:14 -0800, Paul Winkler
wrote:
No, its a primitive cabinet simulator. Just model reflections off the back
wall. The LP represents absorbion in the fiberous stuff inside. I read
about the technique somewhere... music-dsp probably.
i do think that the in-cab echo is part of the impulse response
if you have a convolver ('imp_1199', 'unmatched') following the
shaper summation node. it sounds to me like it is fwiw.
how many split bands do you think we'll need?
nice ascii, btw.
tim
For the
record my current guess for an amp process is:
.-> bandpass -> shaper -. .--- LP <--.
| | v |
input -> EQ? -+-> bandpass -> shaper -+-> delay -> EQ ---> output
| |
`-> bandpass -> shaper -'
well, we'd have to try it and see :)
the delay w/ lp in feedback - is that basically
how you make a resonant lopass filter?
5 Nov
5 Nov
4:57 a.m.
On Tue, Nov 05, 2002 at 01:56:21 +0100, Tim Goetze wrote:
The impulse includes the cabinet response, but the IIR version you
produced I'm not so sure about, I'd have to run the IIR through octave to
find out I guess, but (looking at it simplisticly) I dont think the
response is long enough to include the cabinet effect.
> > For
the record my current guess for an amp process is:
> >
> > .-> bandpass -> shaper -. .--- LP <--.
> > | | v |
> > input -> EQ? -+-> bandpass -> shaper -+-> delay -> EQ --->
output
> > | |
> > `-> bandpass -> shaper -'
>
i do think that the in-cab echo is part of the impulse response
if you have a convolver ('imp_1199', 'unmatched') following the
shaper summation node. it sounds to me like it is fwiw. how many split bands do you think we'll need?
My guess is one for every two octaves. But that is absolutly a guess.
What is the frequency range of a guitar amp? If its more than 8 octaves it
might be neccesary to widen the bandpasses past two octaves, or maybe the
top end can just be lightly shaped together.
Who'd think one small wooden box could be so complicated ;)
- Steve
2:37 p.m.
Steve Harris wrote:
8 octaves would be something like 40 - 10240 Hz, that should do
it.
The impulse includes the cabinet response, but the IIR
version you
produced I'm not so sure about, I'd have to run the IIR through octave to
find out I guess, but (looking at it simplisticly) I dont think the
response is long enough to include the cabinet effect.
have you looked at plots of the impulses yet? you can see an echo
pattern in them afaict.
how many split
bands do you think we'll need?
My guess is one for every two octaves. But that is absolutly a guess.
What is the frequency range of a guitar amp? If its more than 8 octaves it
might be neccesary to widen the bandpasses past two octaves, or maybe the
top end can just be lightly shaped together. Who'd think one small wooden box could be so
complicated ;)
you wouldn't call it small if i hired you to be my roady. ;)
tim
4:05 p.m.
On Tue, Nov 05, 2002 at 07:32:21PM +0100, Tim Goetze wrote:
8 octaves would be something like 40 - 10240 Hz, that
should do
it.
Truly godawful image quality, but this may be of interest -
it's a plot of a celestion "Greenback" 12" gguitar speaker
in an anechoic chamber:
http://home3.netcarrier.com/~lxh2/greenb.gif
here's some others, which I don't know how they
were measured, but at least you can read the
Y-axis scale:
http://home3.netcarrier.com/~lxh2/celvint.gif
--
Paul Winkler
http://www.slinkp.com
"Welcome to Muppet Labs, where the future is made - today!"
6:57 p.m.
On Tue, Nov 05, 2002 at 12:01:57 -0800, Paul Winkler wrote:
On Tue, Nov 05, 2002 at 07:32:21PM +0100, Tim Goetze
wrote:
Looks to me like there is still useful infrmation at 20k, luckily it looks
like the multiband shaping was a dead end (though it may still make for
interesting effects, the cheby attempt did).
Unluckily it looks like to do a good job we need to tap those 100+ volt
signals :(
- Steve
8 octaves would be something like 40 - 10240 Hz,
that should do
it.
Truly godawful image quality, but this may be of interest -
it's a plot of a celestion "Greenback" 12" gguitar speaker
in an anechoic chamber:
http://home3.netcarrier.com/~lxh2/greenb.gif
7:44 p.m.
On Tue, Nov 05, 2002 at 10:54:02PM +0000, Steve Harris wrote:
Unluckily it looks like to do a good job we need to
tap those 100+ volt
signals :(
I'm pretty handy with a soldering iron, if anyone has a clue how this
could be done. If I remember my moron-level analog electronics,
couldn't I just use a big blocking capacitor?
like:
input stuff -> (tube) --> (tube) --> power amp stuff
^
|
= large electrolytic capacitor rated at e.g. 220V
|
positive end of tap
That should take care of teh high DC voltage, IIRC.
Of course, I have no idea what kind of signal level
you'd have at that point.
--
Paul Winkler
http://www.slinkp.com
"Welcome to Muppet Labs, where the future is made - today!"
6 Nov
6 Nov
5:49 a.m.
Paul Winkler wrote:
> input stuff -> (tube) --> (tube) --> power amp stuff
> ^
> |
> = large electrolytic capacitor rated at e.g. 220V
> |
> positive end of tap
> That should take care of teh high DC voltage, IIRC.
> Of course, I have no idea what kind of signal level
> you'd have at that point.
That´s exactly the way to go, though I wouldn´t
use an electrolytic
capacitor. A suitable buffer is described in oscope-1.7.tar.gz by Tim
Witham (twitham(a)quiknet.com).
The signal level, at least after the first amp stages, shouldn´t exceed
some volts.
I´ve read this thread from the beginning, but as I know next to nothing
about DSP coding, I didn´t jump in. But now as it comes to hardware,
which is my domain, I want to offer my help. Please feel free to contact
me.
Greetings
Klaus
--
Dr. Klaus Kosten
Am Ginsterberg 13
D-52477 Alsdorf
7989
days inactive
7992
days old
linux-audio-dev@lists.linuxaudio.org
19 comments
5 participants
participants (5)
-
Klaus Kosten -
Mark Knecht -
Paul Winkler -
Steve Harris -
Tim Goetze