Re: Deconvolution
- From: "A. Kumar" <akumar.elex@xxxxxxxxxxxx>
- Date: Sat, 11 Mar 2006 11:20:15 -0500
Mike Parham wrote:
I wonder whether anybody would be kind enough to comment on the
sanity of the following:
I have a signal S1 corrupted by (periodic) noise and another
signal,
S2, I hope containg the same signal as the first but corrupted by a
delayed version of the noise. I want to remove the noise and
recover
the signal.
I have tried delaying S1, subtracting S1 and S2 and then
deconvolving
the difference signal with the presumably known(?) transfer
function
(like h = [1 0 0 0 0 -1] with fractional) delay set appropriately).
On synthetic data with it seems to work but in practice my results
are not sensible.
I have tried various deconvolution approaches:
1) deconv (linear deconvolution)
2) IFFT(DFT S1 / DFT h)
3) convolution matrix and pinv based approaches (circular and
linear)
including using the SVD adjustment parameter
Any suggestions for modifications / other approaches I could try or
references to papers/websites would be appreciated. Or is this
method
doomed?
Thanks,
Mike
(1) You may like performing averaging operations i.e noise will
average out to zero something like h[n]=[.5,.5]
or (2) you may consider performing cross correlation between
corrupted noise and the signal. You will get back your signal
(provided original signal is known)
(3) you may consider using wavelet transform for denoising either S1
or S2 independently.
A. Kumar
.
- References:
- Deconvolution
- From: Mike Parham
- Deconvolution
- Prev by Date: Re: Matlab too expensive?
- Next by Date: System of Transcendentals Equations using fsolve
- Previous by thread: Deconvolution
- Next by thread: deconvolution
- Index(es):
Relevant Pages
|
