On 2013-05-01, Dave Hunt wrote:
Dolby A (etc.) decode. Would it not be possible to to do this with
convolution ?? Find a working unit, record its impulse response, use
that in one of the many convolution reverb/filter plug-ins.
Unfortunately this is not an option. Convolution can be used to model
any system that is linear and time invariant, that is, sums and constant
multiples of inputs, even if shifted arbitrarily in time, lead to sums,
multiplications and equal shifts, of the resulting outputs. This does
not hold for compressors or expanders, including multiband ones like the
compander architecture in the various kinds of Dolby NR: even if in very
short term they try to behave roughly linearly so as not to add audible
nonlinear distortion and they don't have time variance like e.g. a
tremolo effect, most decidedly even their short term frequency response
varies and addition and multiplication shift the signals over the
different knees of the compander which is a visible nonlinearity.
Obviously the Dolby reference level would have to be taken into account.
Every time there is a reference level in a system that actually impacts
how it operates, the system is guaranteed to be nonlinear, because
otherwise you could freely multiply the signal by some number a before
the system and by 1/a after the system without changing how the system
sounds.
LTI systems are a broad and useful class of systems, with a nice theory
and beautiful computational properties, but they don't cover all of
audio signal processing by a long shot. The systems you can simulate via
convolution include constant gain, filters/equalisers with constant
settings, echo, delay, reverb and all of their combinations. Linear but
not time invariant systems include chorus, phasing, flanging, wah,
tremolo, vibrato and like things, so they are also out for convolution,
unless you continuously change the impulse response you're convolving
with (usually a bad idea because convolver topology is heavily optimized
for constant coefficients). Then stuff that is decidedly nonlinear is
doubly out: dynamics processing, fuzz/distortion, anything of that sort.
--
Sampo Syreeni, aka decoy - [email protected], http://decoy.iki.fi/front
+358-50-5756111, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2
_______________________________________________
Sursound mailing list
[email protected]
https://mail.music.vt.edu/mailman/listinfo/sursound
