Forums

Deconvolution question

Started by Les Cargill May 9, 2009
Suppose I have an acoustic guitar with a (relatively good-sounding)
peizo pickup (a K&K, for those interested).

I can mic the guitar and record the peizo simultaneously. But when
I deconvolve the miked signal against the peizo, the resulting
convolution signature isn't very coherent. Most convolution
signatures of this nature have a "spike" where the dominant
bucket is - this has just a "hump" or two.

I expect that the peizo is simply massively nonlinear with
respect to the microphone, but what I am not familiar with
is how pure phase transforms affect convolution, as opposed
to transforms more akin to sequences of allpass filters...
in "series" (?) Words fail me a bit here - my presumption
is that the convolution signature is just a sequence of
n allpass filters, each of time constant n*Fs .

What exactly am I looking at here?

Also, the "impulse" used is probably not very broadband. Still,
the convolution signature applied to the peizo doesn't do a
bad job of sounding like the miked channel, if you filter out
the repeating patterns in the signal.

--
Les Cargill
On May 9, 2:17&#2013266080;pm, Les Cargill <lcarg...@cfl.rr.com> wrote:
> Suppose I have an acoustic guitar with a (relatively good-sounding) > peizo pickup (a K&K, for those interested). > > I can mic the guitar and record the peizo simultaneously. But when > I deconvolve the miked signal against the peizo, the resulting > convolution signature isn't very coherent. Most convolution > signatures of this nature have a "spike" where the dominant > bucket is - this has just a "hump" or two. > > I expect that the peizo is simply massively nonlinear with > respect to the microphone, but what I am not familiar with > is how pure phase transforms affect convolution, as opposed > to transforms more akin to sequences of allpass filters... > in "series" (?) Words fail me a bit here - my presumption > is that the convolution signature is just a sequence of > n allpass filters, each of time constant n*Fs . > > What exactly am I looking at here? > > Also, the "impulse" used is probably not very broadband. Still, > the convolution signature applied to the peizo doesn't do a > bad job of sounding like the miked channel, if you filter out > the repeating patterns in the signal. > > -- > Les Cargill
How do you implement "deconvolution" exactly? The devil is in the details. At any rate, I suspect that you the system that you are trying to deconvolve has zeros in the frequency domain, hence straightforward deconvolution will give a very poor result. Julius
Les Cargill wrote:
> > Suppose I have an acoustic guitar with a (relatively good-sounding) > peizo pickup (a K&K, for those interested). > > I can mic the guitar and record the peizo simultaneously. But when > I deconvolve the miked signal against the peizo, the resulting > convolution signature isn't very coherent. Most convolution > signatures of this nature have a "spike" where the dominant > bucket is - this has just a "hump" or two. > > I expect that the peizo is simply massively nonlinear with > respect to the microphone, but what I am not familiar with > is how pure phase transforms affect convolution, as opposed > to transforms more akin to sequences of allpass filters... > in "series" (?) Words fail me a bit here - my presumption > is that the convolution signature is just a sequence of > n allpass filters, each of time constant n*Fs . > > What exactly am I looking at here? > > Also, the "impulse" used is probably not very broadband. Still, > the convolution signature applied to the peizo doesn't do a > bad job of sounding like the miked channel, if you filter out > the repeating patterns in the signal.
Do a little groundwork with http://www.dspguide.com/ch17/2.htm. (Ignore the first set of figures. They go with text on the previous page.) Jerry -- Engineering is the art of making what you want from things you can get. &#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;
julius wrote:
> On May 9, 2:17 pm, Les Cargill <lcarg...@cfl.rr.com> wrote: >> Suppose I have an acoustic guitar with a (relatively good-sounding) >> peizo pickup (a K&K, for those interested). >> >> I can mic the guitar and record the peizo simultaneously. But when >> I deconvolve the miked signal against the peizo, the resulting >> convolution signature isn't very coherent. Most convolution >> signatures of this nature have a "spike" where the dominant >> bucket is - this has just a "hump" or two. >> >> I expect that the peizo is simply massively nonlinear with >> respect to the microphone, but what I am not familiar with >> is how pure phase transforms affect convolution, as opposed >> to transforms more akin to sequences of allpass filters... >> in "series" (?) Words fail me a bit here - my presumption >> is that the convolution signature is just a sequence of >> n allpass filters, each of time constant n*Fs . >> >> What exactly am I looking at here? >> >> Also, the "impulse" used is probably not very broadband. Still, >> the convolution signature applied to the peizo doesn't do a >> bad job of sounding like the miked channel, if you filter out >> the repeating patterns in the signal. >> >> -- >> Les Cargill > > How do you implement "deconvolution" exactly?
I use a software package that calculates the deconvolution. This process is opaque to me; I don't know what exactly it does. I presume it uses one of the many "cookbook" algorithms I've seen for applying FFT to produce deconvolutions.
> The devil > is in the details. > > At any rate, I suspect that you the system that you are > trying to deconvolve has zeros in the frequency domain, > hence straightforward deconvolution will give a very > poor result. >
I'm not sure how to get this into the frequency domain.
> Julius
-- Les Cargill