26 Feb
2015
26 Feb
'15
11:06 p.m.
Date: Thu, 26 Feb 2015 22:04:39 +0000
From: Fons Adriaensen <fons(a)linuxaudio.org>
To: Cedric Roux <sed(a)free.fr>
On Wed, Feb 25, 2015 at 11:46:37PM +0100, Cedric Roux wrote:
can someone explain to me what this bandwidth
computation means?
There is nothing magical about it, it's just a pragmatic
approximation that results in the 3dB BW (for high + or -
gain) being expressed in octaves.
Expressing the BW in a logarithmic unit (such as octaves)
make sense because the magnitude response of a second order
parametric is symmetric on a log(f) scale but not on a linear
one.
An absolute value in Hz or even a relative one (the same divided
by center frequency) is informative only for small BW, where the
linear distance to the -3dB points left and right would be more
or less equal.
For large BW values (as often used in sound engineering) this
is no longer true. For example for a center frequency of 1 kHz
and two octaves BW the -3dB points are at 500 Hz and 2 kHz.
For four octaves that would be 250 Hz and 4 kHz, giving a -3dB
BW in Hz that is larger than twice the center frequency, which
is somewhat counter-intuitive.
The 'mathematical' way to define the BW of an second order
allpass would be at the +/- 90 degree phase points. This also
corresponds to the -3 dB points for an infinite notch, as used
by M & R. But keeping this value fixed in a parametric results
in curves that are not even symmetric for + and - gains. The
'bumps' would be much wider than the 'notches'. The sqrt (gain)
factor restores this symmetry. But using this means that there
is no longer any simple relation to what looks like bandwidth
on a frequency response plot and the actual 'mathematical'
value.
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
----- End forwarded message -----
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
27 Feb
27 Feb
11:20 p.m.
On 02/26/2015 11:06 PM, Fons Adriaensen wrote:
Date: Thu, 26 Feb 2015 22:04:39 +0000
From: Fons Adriaensen <fons(a)linuxaudio.org>
To: Cedric Roux <sed(a)free.fr>
On Wed, Feb 25, 2015 at 11:46:37PM +0100, Cedric Roux wrote:
Thanks for explanations Fons.
I tried gamma = = sinh(log(2)/2 * bw * w0 / sin(w0)) * sin(w0)
as found in [1].
It sets gain(bandfreqs) = gain/2 (it's not the -3dB thing).
That works except for high f0 and high bandwidth.
In the worst case it acts on all the spectrum, turning itself
into a gain control. Strange. Your formula behaves cleaner
even if not 'mathematically' exact.
Now I have a minor issue with your smoothing techniques.
They depend on the block size.
Initializing everything, setting block size to 1 sample
(why not? we're in 2015!) and printing s1 and s2 for
the processing of a few samples will result in a list of values.
Setting block size to 16, 32, 48 samples gives different
values (after a while they agree though, which is expected).
Well, it's no big deal, this is an eq, no sane person
will ever use a block size of 1 sample (or anything
below 32 samples) and for all usual block sizes (32, 64,
128, ..., 2^n) the result is the same.
So forget about it.
I would just want to know why you do:
k = (len > 48) ? 32 : len;
why 48? why not 32? What's your intention there?
Regards,
Cédric.
[1] http://www.musicdsp.org/files/EQ-Coefficients.pdf
can someone explain to me what this bandwidth
computation means?
There is nothing magical about it, it's just a pragmatic
approximation that results in the 3dB BW (for high + or -
gain) being expressed in octaves.
11:44 p.m.
On Fri, Feb 27, 2015 at 11:18:04PM +0100, Cedric Roux wrote:
I would just want to know why you do:
k = (len > 48) ? 32 : len;
why 48? why not 32? What's your intention there?
It avoids having a very short block at the end - it will be
absorbed by the previous one up to a maximum of 48 frames.
For example 100 = 32 + 32 + 32 + 4 becomes 32 + 32 + 36.
This plugin is very old, in more recent software (the
zita-* series) control values are evaluated at exactly
every N frames (where N depends on the algorithm being
controlled).
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
3553
days inactive
3554
days old
linux-audio-dev@lists.linuxaudio.org
2 comments
2 participants
participants (2)
-
Cedric Roux -
Fons Adriaensen