27 Mar
2010
27 Mar
'10
3:05 a.m.
Throughout the decades, I've read various engineers and producers waxing poetically
and rapturously about the Neve console filters and how great they sound.
Filters are filters, and there must be some way to measure their their frequency
characteristics and emulate it in software. Has anyone done that? In LADSPA on Linux?
-ken
27 Mar
27 Mar
8:18 a.m.
On Fri, Mar 26, 2010 at 11:04:00PM -0700, Ken Restivo wrote:
Throughout the decades, I've read various
engineers and
producers waxing poetically and rapturously about the Neve
console filters and how great they sound.
Same about Helios, Cadac, Midas, ...
I worked on big Neve consoles for years, and sure I do
like the filters. But see below.
Filters are filters, and there must be some way to
measure
their their frequency characteristics and emulate it in
software. Has anyone done that? In LADSPA on Linux?
We're close to myths and pseudoscience here.
Filters must not add distortion, noise etc. They only change
the frequency response. Now classic analog filters used in
audio consoles are (with maybe some rare exceptions) all of
the 1st and 2nd order minimum phase type. Which means that
(given the required controls and ranges) they all do the same
thing.
What makes them different is how they *feel* - the relation
between feeling the knobs move and the actual response.
For shelf filters 'gain' and 'frequency' will interact in
some way, and for parametrics 'gain' and 'bandwidth' will.
There are different ways to define these interactions, and
most engineers will prefer one or the other.
Unless Neve (or any of the others) are doing something really
exceptional (and then we'd know it), I'm pretty sure that if
you take e.g. the Neve circuit, replace the potentiometers
with some other brand that have either more or less friction,
change the type, size and color of the knobs, and modify the
layout a bit, nobody would recognise it as 'the Neve filter'.
Neve are still in business, you can buy the equalisers. If
you do it would be interesting to measure them, but I'd be
surprised if anything special would turn up.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !
9:53 a.m.
I read something in tape-op about some "golden eared" guy who detected
something wrong with a neve desk IIRC and when they analysed it, it
turned out there was some problem with the freq response above 60K or
some other equally unbelievable figure. From what I read they are
supposed to have a really transparent sound (whatever that means). I
doubt something similar could be done with digital as the placebo
effect is completely lost!
On 27/03/2010, fons(a)kokkinizita.net <fons(a)kokkinizita.net> wrote:
On Fri, Mar 26, 2010 at 11:04:00PM -0700, Ken Restivo
wrote:
--
Sent from my mobile device
Throughout the decades, I've read various
engineers and
producers waxing poetically and rapturously about the Neve
console filters and how great they sound.
Same about Helios, Cadac, Midas, ...
I worked on big Neve consoles for years, and sure I do
like the filters. But see below.
Filters are filters, and there must be some way
to measure
their their frequency characteristics and emulate it in
software. Has anyone done that? In LADSPA on Linux?
We're close to myths and pseudoscience here.
Filters must not add distortion, noise etc. They only change
the frequency response. Now classic analog filters used in
audio consoles are (with maybe some rare exceptions) all of
the 1st and 2nd order minimum phase type. Which means that
(given the required controls and ranges) they all do the same
thing.
What makes them different is how they *feel* - the relation
between feeling the knobs move and the actual response.
For shelf filters 'gain' and 'frequency' will interact in
some way, and for parametrics 'gain' and 'bandwidth' will.
There are different ways to define these interactions, and
most engineers will prefer one or the other.
Unless Neve (or any of the others) are doing something really
exceptional (and then we'd know it), I'm pretty sure that if
you take e.g. the Neve circuit, replace the potentiometers
with some other brand that have either more or less friction,
change the type, size and color of the knobs, and modify the
layout a bit, nobody would recognise it as 'the Neve filter'.
Neve are still in business, you can buy the equalisers. If
you do it would be interesting to measure them, but I'd be
surprised if anything special would turn up.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !
_______________________________________________
Linux-audio-user mailing list
Linux-audio-user(a)lists.linuxaudio.org
http://lists.linuxaudio.org/listinfo/linux-audio-user
11:04 a.m.
On Sat, Mar 27, 2010 at 12:53:22PM +0000, James Stone wrote:
I read something in tape-op about some "golden
eared" guy who detected
something wrong with a neve desk IIRC and when they analysed it, it
turned out there was some problem with the freq response above 60K or
some other equally unbelievable figure. From what I read they are
supposed to have a really transparent sound (whatever that means).
Whatever that means, indeed.
A classic way for analog circuits to fail is to have some
oscillation above the audio range - up to several MHz is
possible - and that can affect the sound in very sublte
ways, and depending on the signal itself. More often than
not such things depend on how the circuit is connected,
on wiring capacitances, power supply impedances, etc.
Problems like this can be very different on production
hardware and laboratory prototypes, and are not always
revealed by traditional measurement methods. It wouldn't
be the first time for such a defect to remain unnoticed
for years. Until someone hits exactly the type of audio
signal that gets affected in a percetible way, and has
the ears to detect it. But still these are defects,
nothing magical about them.
Also in the era when manufacturers such as Neve made
their name, there were *lots* of badly designed filter
circuits in use. So if Neve or one of the others made
something that was 'just OK', it would get noticed and
create a myth, even if objectively it was just a good
quality circuit without problems.
I doubt something similar could be done with digital
as the placebo
effect is completely lost!
Placebo effects and nonsense exist in the digital world
as well, and are even harder to debunk.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !
11:45 a.m.
On Sat, 27 Mar 2010 14:59:53 +0100
fons(a)kokkinizita.net wrote:
What! Are you telling me that going round the edge of a CD with a felt
tip pen *doesn't* improve the 'clarity'? :D
--
Will J Godfrey
http://www.musically.me.uk
Say you have a poem and I have a tune.
Exchange them and we can both have a poem, a tune, and a song.
On Sat, Mar 27, 2010 at 12:53:22PM +0000, James Stone
wrote:
<snip>
I doubt
something similar could be done with digital as the placebo
effect is completely lost!
Placebo effects and nonsense exist in the digital world
as well, and are even harder to debunk.
Ciao,
1:19 p.m.
On Sat, Mar 27, 2010 at 02:45:18PM +0000, Folderol wrote:
What! Are you telling me that going round the edge of
a CD with a felt
tip pen *doesn't* improve the 'clarity'? :D
Depends on the price of the felt tip pen.
Regarding the 'feel' of filters, I've put up some examples
on <http://www.kokkinizita.net/linuxaudio/paramfilt.html>.
These are three examples of a parametric filter set to
1 kHz, fixed 'bandwidth' and gain between -15 and +15 dB
in steps of 5 dB. The curves for +15 dB are identical for
all three.
If you compare Type 1 and Type 2, it's easy to see that
the two would 'feel' quite different if you turn the gain
knob. Most filters you'll see in practice will be somewhere
in between these two. For example my Ladspa filter plugin
is such a compromise, visually somewhat closer to 1 than
to 2.
Type 3 looks quite odd, but it is in a sense the most
'natural' of all three: it corresponds to a variable amount
of a *fixed* bandwidth (i.e. independent of gain) 2nd order
resonance being added or subtracted. Again this 'feels'
different.
Now the DSP code used to produce these curves is identical
for all. It has three internal parameters, c1, c2, g, and
the only difference between the three forms is in the way
that the three 'user' paramters Frequency, Bandwidth and
Gain are mapped to the internal ones. Given a certain curve
(i.e. values for c1, c2, g) all three can produce it.
So this is in fact three time the *same* filter, just
'presented' differently.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !
3:49 p.m.
On Sat, Mar 27, 2010 at 05:15:18PM +0100, fons(a)kokkinizita.net wrote:
On Sat, Mar 27, 2010 at 02:45:18PM +0000, Folderol
wrote:
None of this really addresses the difference in *sound* of these
supposedly-"magical" filters.
But Google turned this up, which I found interesting:
http://www.playgroundstudio.com/blog/?p=60
Has anyone tried emulating these Neve filters in software using LADSPA?
Anyone want to? (um, hi Fons!)
-ken
What! Are you telling me that going round the
edge of a CD with a felt
tip pen *doesn't* improve the 'clarity'? :D
Depends on the price of the felt tip pen.
Regarding the 'feel' of filters, I've put up some examples
on <http://www.kokkinizita.net/linuxaudio/paramfilt.html>.
These are three examples of a parametric filter set to
1 kHz, fixed 'bandwidth' and gain between -15 and +15 dB
in steps of 5 dB. The curves for +15 dB are identical for
all three.
If you compare Type 1 and Type 2, it's easy to see that
the two would 'feel' quite different if you turn the gain
knob. Most filters you'll see in practice will be somewhere
in between these two. For example my Ladspa filter plugin
is such a compromise, visually somewhat closer to 1 than
to 2.
Type 3 looks quite odd, but it is in a sense the most
'natural' of all three: it corresponds to a variable amount
of a *fixed* bandwidth (i.e. independent of gain) 2nd order
resonance being added or subtracted. Again this 'feels'
different.
Now the DSP code used to produce these curves is identical
for all. It has three internal parameters, c1, c2, g, and
the only difference between the three forms is in the way
that the three 'user' paramters Frequency, Bandwidth and
Gain are mapped to the internal ones. Given a certain curve
(i.e. values for c1, c2, g) all three can produce it.
So this is in fact three time the *same* filter, just
'presented' differently.
6:20 p.m.
On Sat, Mar 27, 2010 at 11:47:04AM -0700, Ken Restivo wrote:
None of this really addresses the difference in
*sound*
of these supposedly-"magical" filters.
Because there is none. The only difference is the range
of sounds available by changing a single parameter, e.g.
gain.
The difference between the SSL and Neve parametrics
is exactly the one I've described. Again, each of the
two can do whatever the other can. The only difference
is that instead of turning one knob to change between
curves in the same picture you may have to turn two.
It would be trivial to make e.g. my Ladpsa plugin
switch between the two. In fact the original version
(a Jack app, not published) provided 4 modes for each
section, one of them 'SSL' style and another 'Neve'.
The Neve shelf with the dip before it rises is rather
nice. But again it doesn't take a special filter to do
that. A normal shelf (+ gain) plus a parametric at the
right frequency (- gain) will have exactly the same
effect. If you *know* that you want that kind of curve,
any parametric can do it. Neve provides it on a single
knob, which may be nice in some cases, and a nuisance
in others.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !
9:57 p.m.
On Sat, Mar 27, 2010 at 11:47:04AM -0700, Ken Restivo wrote:
I've been looking a bit deeper into the peculiar shelf
filter shown on that page.
What is shown there is actually a second order shelf
filter - the sum of the input + a second order highpass
with some gain.
This correspons to the blue curve, the one that 'dips'
the most before starting to rise. The other curves are
the result if you move the two sections of the highpass
apart in frequency, gradually turning it into 1st order
for the part that matters - this is the yellow curve.
Now what is the origin of this 'dip' ? For a 2nd order
highpass the part below the cutoff frequency has a phase
that tends towards 180 degrees as frequency goes down and
the slope of the filter approaches 12 dB/oct. At the point
where the highpass_with_gain curve intersects the 0dB line,
the phase shift has not yet reached 180 degrees, but it is
well beyond 90 degrees so the real part is in antiphase
with input signal it is added to, and that causes the 'dip'.
In an analog implementation this would be difficult to
avoid. So this 'dip' may well be an unwanted side effect
rather than a wanted feature, with the real feature being
the 2nd order nature of the filter giving it a slope that
is almost double the normal value for a shelf filter.
(look at the blue line: 11 dB difference from 2 to 4 kHz).
This is relatively easy to implement, maybe I'll do it
one of these days.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !
5154
days inactive
5154
days old
linux-audio-user@lists.linuxaudio.org
8 comments
4 participants
participants (4)
-
Folderol -
fons＠kokkinizita.net -
James Stone -
Ken Restivo