Spectral subtraction

Hi,

I would like to know if you think it's possible to implement a noise
reduction system in every receivers with spectral subtraction method (for
example could it reduce noise in a signal with a bandwith of 10MHz at a
frequency of 500MHz)?

Thank you

Thom

"thom" <soniceric@hotmail.com> wrote in message 
news:svKdna87uJkuTM3ZnZ2dnUVZ_tmdnZ2d@giganews.com...
> Hi,
>
> I would like to know if you think it's possible to implement a noise
> reduction system in every receivers with spectral subtraction method (for
> example could it reduce noise in a signal with a bandwith of 10MHz at a
> frequency of 500MHz)?
>
> Thank you
>
> Thom

Hello Thom,

Basically all receivers employ noise reduction via spectral subtraction. The 
preselect and IF filters do exactly this. Now if the noise is in your 
channel of interest  you need to do more.

Clay

>Hello Thom,
>
>Basically all receivers employ noise reduction via spectral subtraction.
The 
>preselect and IF filters do exactly this. Now if the noise is in your 
>channel of interest  you need to do more.
>
>Clay


Talking about "spectral subtraction", I was thinking about the methods to
reduce background noise in audio signals (by Fast Fourier Transform of a
signal frame, and then estimation of noise power and subtraction in the
frequency domain). The problem is that I think the signal has to be
quasi-stationary in the frame... So I'm not sure that we can have frames
where the signal is quasi-stationary if the signal frequency is too
high...

What do you think about that?

If you have other methods to reduce noise in the channel of interest, I'm
also interested!


Thank you

Thom




  

>
> What do you think about that?
>

Let an analysis window slide over your signal and calculate the mean
and variance of the window each time you move the window. Plot
the results and you will be able to see how stationary your signal
is and determine the width of your analysis window accordingly.

Clay S. Turner wrote:
> "thom" <soniceric@hotmail.com> wrote in message
> news:svKdna87uJkuTM3ZnZ2dnUVZ_tmdnZ2d@giganews.com...
> > Hi,
> >
> > I would like to know if you think it's possible to implement a noise
> > reduction system in every receivers with spectral subtraction method (for
> > example could it reduce noise in a signal with a bandwith of 10MHz at a
> > frequency of 500MHz)?
> >
> > Thank you
> >
> > Thom
>
> Hello Thom,
>
> Basically all receivers employ noise reduction via spectral subtraction. The
> preselect and IF filters do exactly this.

Hmmm.... I may be splitting hairs here, but I think this at best is a
gross
oversimplification; at worst it is plain wrong.

A filter does not *subtract* noise from a frequency band. It
*mutliplies*
the noise inside that band wit a factor H(w) << 1. The difference is
that
with multiplication, you don't need to know the amplitude of the
noise, whaile if you want to subtract in spectrum domain, you both
need to know the amplitude and the phase, or you will de facto
*add* noise to your signal.

So spectrum subtraction is not a very good way to do things. 

Rune

"John" <john@jnho.hnjo.invalid.com> wrote in message 
news:e2rdld$7il$1@newsbin.cybercity.dk...
> >
>> What do you think about that?
>>
>
> Let an analysis window slide over your signal and calculate the mean
> and variance of the window each time you move the window. Plot
> the results and you will be able to see how stationary your signal
> is and determine the width of your analysis window accordingly.
>
Also can have a look at 
http://www.ulg.ac.be/telecom/teaching/notes/total1/elen008/node77_id.html
with particular attention paid to the autocorrelation function.
Best of luck - Mike

So if spectral subtraction isn't the right solution, do you know other
methods to reduce white noise in a "broadband" signal?? (BW of several
MHz)

Rune Allnor wrote:

> Clay S. Turner wrote:
> > "thom" <soniceric@hotmail.com> wrote in message
> > news:svKdna87uJkuTM3ZnZ2dnUVZ_tmdnZ2d@giganews.com...
> > > Hi,
> > >
> > > I would like to know if you think it's possible to implement a noise
> > > reduction system in every receivers with spectral subtraction method (for
> > > example could it reduce noise in a signal with a bandwith of 10MHz at a
> > > frequency of 500MHz)?
> > >
> > > Thank you
> > >
> > > Thom
> >
> > Hello Thom,
> >
> > Basically all receivers employ noise reduction via spectral subtraction. The
> > preselect and IF filters do exactly this.
>
> Hmmm.... I may be splitting hairs here, but I think this at best is a
> gross
> oversimplification; at worst it is plain wrong.
>
> A filter does not *subtract* noise from a frequency band. It
> *mutliplies*
> the noise inside that band wit a factor H(w) << 1. The difference is
> that
> with multiplication, you don't need to know the amplitude of the
> noise, whaile if you want to subtract in spectrum domain, you both
> need to know the amplitude and the phase, or you will de facto
> *add* noise to your signal.

The process called "spectral subtraction" described in this article

http://cslu.ece.ogi.edu/nsel/wan_manuscript/node10.html

works slightly differently. The estimated noise power spectrum is
subtracted from the measured signal power spectrum, and then converted
back to a DFT spectrum by using the phase of the original DFT used to
compute the signal power spectrum.

I might be wrong, but doesn't this method imply a constant gain in case
the noise is white? That doesn't seem like a very effective noise
reduction system to me ... but then, I'm probably missing something.

> So spectrum subtraction is not a very good way to do things.

>From the OPs question, a simple filter might already do the job.
Depending on the signal, more sophisticated methods could be used.

Yes with a filter, you can reduce noise which is outside the band of
interest... But what about reducing the in-band noise? I was studying
adaptive filtering, but it seems to work if you have a reference of the
desired signal (this is not my case...). Someone knows other techniques?

thom wrote:

> Yes with a filter, you can reduce noise which is outside the band of
> interest... But what about reducing the in-band noise? I was studying
> adaptive filtering, but it seems to work if you have a reference of the
> desired signal (this is not my case...). Someone knows other techniques?

Adaptive filters also work if the correlation length between the signal
and noise differ. Two cases are possible:

- non-correlated noise, correlated signal (eg. sinusoidal signal in
white noise)
- long-term correlated noise, short-term correlated signal (eg. music
signal with sinusoidal interference)

In those cases, you can use an adaptive n-step predictor with
appropriate prediction gap n (must lie in between the two correlation
lengths).