Wiener Filter for noise reduction (URGENT HELP REQUIRED)

EDAboard.com Forum Index -> Digital Signal Processing -> Wiener Filter for noise reduction (URGENT HELP REQUIRED)

Author

Message

eksitara

Joined: 13 Feb 2008
Posts: 1

12 Mar 2008 16:20 Wiener Filter for noise reduction (URGENT HELP REQUIRED)

Hi,

I am doing my project in noise cancellation and i am implementing it using matlab.I have recorded my noisy speech and simultaneously recorded the noise using different recording device.I am making a FIR filter which will remove or reduce the noise.

The theory says that if

x(n)=d(n)+v(n)

d(n)=desired signal
v(n)=noise

Now in order to calculate the coefficients for the filter which will produce a noise v'(n) and when when added to x(n) will remove the noise and leave the desired signal.

To calculate the coeffiecients of the filter

b=Rdx/Rxx

Where Rdx is the cross correlation estimate between d(n) and x(n) and Rxx is auto correlation between x(n)

Problem: (1) How do i cross correlate d(n) and x(n) i know it is xcorr function but how do i obtain only d(n) signal?
(2) How should i use the filter function to implement my coefficients on the noisey signal to get v'(n)?

[/url]

Google
AdSense
Google Adsense

12 Mar 2008 16:20 Ads

FvM

Joined: 22 Jan 2008
Posts: 5160
Helped: 767
Location: Bochum, Germany

12 Mar 2008 19:52 Re: Wiener Filter for noise reduction (URGENT HELP REQUIRED)

Code:

calculate the coefficients for the filter which will produce a noise v'(n) and when added to x(n) will remove the noise and leave the desired signal

That's basically a misunderstanding of the optimum filter concept. In the usual case, when you have uncorrelated noise and an "orthogonal" signal, a filter only can reduce spectral components of noise outside the signal band. Wiener filter theory is a rule to calculate this filter, no matter how big or small it's effect may be. With a speech signal, it probably isn't very effective. Substracting a noise component, as you imagine, would only be possible with a correlated (e. g. periodical noise) or noise correlated to the signal, e. g. nonlinear distortions.

mathuranathan

Joined: 11 Sep 2006
Posts: 247
Helped: 29
Location: India

14 Mar 2008 8:11 Re: Wiener Filter for noise reduction (URGENT HELP REQUIRED)

Step to cross-correlate:
* First sample ur desired signals d(n) and x(n).
* Get the sampled values and use xcorr function in signal processing tool box for cross-correlation.
(Refer : http://www.mathworks.com/access/helpdesk/help/toolbox/signal/ )

for "filter" usage refer to
http://www.mathworks.com/access/helpdesk/help/toolbox/signal/index.html?/access/helpdesk/help/toolbox/signal/filter.html&http://www.mathworks.com/cgi-bin/texis/webinator/search/?db=MSS&prox=page&rorder=750&rprox=750&rdfreq=500&rwfreq=500&rlead=250&sufs=0&order=r&is_summary_on=1&ResultCount=10&query=filter&submitButtonName=Search