Highly Non-Stationary Signals for Adaptive Filters

Hello,

I encountered a problem with Noise Cancellation.  

The interfering signal that I am trying to cancel goes on and off really
fast (noise of a switch).  I have never quantified how long it turns on and
off but it it is less than 1 ms or so. 

I am attempting to cancel its noise by applying an adaptive filter and my
application works for stationary signals such as band-limited white
Gaussian noise, but it doesn't work for this problem.

Assuming that the noise of the interfering signal is the same when it turns
on and off, can I turn off the adaptation when the switch is off, and turn
on the adaptation when the switch turns on so that my filter doesn't
misalign and continue to find the MMSE.

Thanks

On 01/19/2011 12:23 PM, pacman101 wrote:
> Hello,
>
> I encountered a problem with Noise Cancellation.
>
> The interfering signal that I am trying to cancel goes on and off really
> fast (noise of a switch).  I have never quantified how long it turns on and
> off but it it is less than 1 ms or so.
>
> I am attempting to cancel its noise by applying an adaptive filter and my
> application works for stationary signals such as band-limited white
> Gaussian noise, but it doesn't work for this problem.
>
> Assuming that the noise of the interfering signal is the same when it turns
> on and off, can I turn off the adaptation when the switch is off, and turn
> on the adaptation when the switch turns on so that my filter doesn't
> misalign and continue to find the MMSE.

So is this noise that is present whenever the switch is on, but is not 
present when the switch is off, or is it a noise spike that occurs 
whenever the switch changes state?

In the first case, if you can somehow know whether the noise is present 
then yes, turning adaptation off when the noise isn't there is probably 
a good idea.  In the second, it would seem that the noise duration is 
too short to filter out in a 'normal way' (i.e. with a linear filter) 
and something like a noise blanker may be better.

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

Yes the switch generates noise while it is on, not when it changes state.

Is there a benefit to oversampling as well?  If I were to oversample the
signal much more, do I reach convergence much faster from a time domain
perspective (not number of samples).

What exactly is a noise blanker?

>On 01/19/2011 12:23 PM, pacman101 wrote:
>> Hello,
>>
>> I encountered a problem with Noise Cancellation.
>>
>> The interfering signal that I am trying to cancel goes on and off
really
>> fast (noise of a switch).  I have never quantified how long it turns on
and
>> off but it it is less than 1 ms or so.
>>
>> I am attempting to cancel its noise by applying an adaptive filter and
my
>> application works for stationary signals such as band-limited white
>> Gaussian noise, but it doesn't work for this problem.
>>
>> Assuming that the noise of the interfering signal is the same when it
turns
>> on and off, can I turn off the adaptation when the switch is off, and
turn
>> on the adaptation when the switch turns on so that my filter doesn't
>> misalign and continue to find the MMSE.
>
>So is this noise that is present whenever the switch is on, but is not 
>present when the switch is off, or is it a noise spike that occurs 
>whenever the switch changes state?
>
>In the first case, if you can somehow know whether the noise is present 
>then yes, turning adaptation off when the noise isn't there is probably 
>a good idea.  In the second, it would seem that the noise duration is 
>too short to filter out in a 'normal way' (i.e. with a linear filter) 
>and something like a noise blanker may be better.
>
>-- 
>
>Tim Wescott
>Wescott Design Services
>http://www.wescottdesign.com
>
>Do you need to implement control loops in software?
>"Applied Control Theory for Embedded Systems" was written for you.
>See details at http://www.wescottdesign.com/actfes/actfes.html
>

On 01/19/2011 01:26 PM, pacman101 wrote:
(top posting fixed)
>> On 01/19/2011 12:23 PM, pacman101 wrote:
>>> Hello,
>>>
>>> I encountered a problem with Noise Cancellation.
>>>
>>> The interfering signal that I am trying to cancel goes on and off
> really
>>> fast (noise of a switch).  I have never quantified how long it turns on
> and
>>> off but it it is less than 1 ms or so.
>>>
>>> I am attempting to cancel its noise by applying an adaptive filter and
> my
>>> application works for stationary signals such as band-limited white
>>> Gaussian noise, but it doesn't work for this problem.
>>>
>>> Assuming that the noise of the interfering signal is the same when it
> turns
>>> on and off, can I turn off the adaptation when the switch is off, and
> turn
>>> on the adaptation when the switch turns on so that my filter doesn't
>>> misalign and continue to find the MMSE.
>>
>> So is this noise that is present whenever the switch is on, but is not
>> present when the switch is off, or is it a noise spike that occurs
>> whenever the switch changes state?
>>
>> In the first case, if you can somehow know whether the noise is present
>> then yes, turning adaptation off when the noise isn't there is probably
>> a good idea.  In the second, it would seem that the noise duration is
>> too short to filter out in a 'normal way' (i.e. with a linear filter)
>> and something like a noise blanker may be better.
>>
 > Yes the switch generates noise while it is on, not when it changes state.
 >
 > Is there a benefit to oversampling as well?  If I were to oversample the
 > signal much more, do I reach convergence much faster from a time domain
 > perspective (not number of samples).

I don't know for sure, but I suspect that you reach a point of sharply 
diminishing returns -- let's hope that someone with more immediate 
experience can answer this.

 > What exactly is a noise blanker?

It's a circuit for reducing noise impulses.  It detects them, and blanks 
a receiver while they're happening.  It doesn't do much good for 
stationary Gaussian noise, but it's good for static discharge.  It's 
also good for Soviet over-the-horizon radar, if you happen to 
inadvertently get transported back in time 30 years.

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

On 1/19/2011 12:23 PM, pacman101 wrote:
> Hello,
>
> I encountered a problem with Noise Cancellation.
>
> The interfering signal that I am trying to cancel goes on and off really
> fast (noise of a switch).  I have never quantified how long it turns on and
> off but it it is less than 1 ms or so.
>
> I am attempting to cancel its noise by applying an adaptive filter and my
> application works for stationary signals such as band-limited white
> Gaussian noise, but it doesn't work for this problem.
>
> Assuming that the noise of the interfering signal is the same when it turns
> on and off, can I turn off the adaptation when the switch is off, and turn
> on the adaptation when the switch turns on so that my filter doesn't
> misalign and continue to find the MMSE.
>
> Thanks

Some of the ideas seem OK but I'm unclear as to the nature of the noise. 
  The adaptive part of a noise canceller will shut off at frequencies 
where there's only random noise because it would only add to random 
noise you'd like to not have in the signal.  So, noise cancellers only 
pass and cancel periodic "noise" like machinery noise.

A line enhancer will do the same thing in a slightly different topology 
with the objective of shutting off random noise and bandpass periodic 
signal.

The former needs a reference for the noise.  The latter doesn't.
The former deals with periodic (having spectral line structure) noise.
The latter deals with random noise in the face of periodic signal.

I don't know if this helps as the parameters of your situation are still 
quite fuzzy.

Fred

On Jan 19, 2:23&#4294967295;pm, "pacman101" <janpac01@n_o_s_p_a_m.yahoo.com>
wrote:
> Hello,
>
> I encountered a problem with Noise Cancellation. &#4294967295;
>
> The interfering signal that I am trying to cancel goes on and off really
> fast (noise of a switch). &#4294967295;I have never quantified how long it turns on and
> off but it it is less than 1 ms or so.
>
> I am attempting to cancel its noise by applying an adaptive filter and my
> application works for stationary signals such as band-limited white
> Gaussian noise, but it doesn't work for this problem.
>
> Assuming that the noise of the interfering signal is the same when it turns
> on and off, can I turn off the adaptation when the switch is off, and turn
> on the adaptation when the switch turns on so that my filter doesn't
> misalign and continue to find the MMSE.
>
> Thanks

As Fred stated, the LMS-based adaptive filter works on signals that
can be correlated. Without knowing the situation you have, I would
suggest you look at sub-band decomposition adaptive-gain filters.
Decompose the signal into sub-bands, determine the SNR for each sub-
band, apply a gain proportional to the SNR for each sub-band, then
reconstruct by adding the sub-bands. This is a non-parametric method
for noise cancellation.

Maurice Givens

Hi, 

That sounds interesting: sub-band decomposition adaptive-gain filters.

I have been doing some searching but I can only find the application to
equalizers.  Could you point a reference where it is applied to noise
cancellation?  That looks really interesting.

I am still trying to quantify the problem.  FYI I am not trying to cancel
Acoustic Noise, but RF noise.  The noise being generated is right on top of
my frequency band.  The switch is part of a system that for some reason
generates RF noise (EMI) when it is on but not when it is off.  

My receive antenna receives the random noise and that's where I came to
this idea.  I could move the antenna, but I can only move it so far that I
still receive the noise.  The receive signal is already really weak and
there is a big difference between the two (about 20-40 dB at different
bands).    
  
>On Jan 19, 2:23=A0pm, "pacman101" <janpac01@n_o_s_p_a_m.yahoo.com>
>wrote:
>> Hello,
>>
>> I encountered a problem with Noise Cancellation. =A0
>>
>> The interfering signal that I am trying to cancel goes on and off
really
>> fast (noise of a switch). =A0I have never quantified how long it turns
on=
> and
>> off but it it is less than 1 ms or so.
>>
>> I am attempting to cancel its noise by applying an adaptive filter and
my
>> application works for stationary signals such as band-limited white
>> Gaussian noise, but it doesn't work for this problem.
>>
>> Assuming that the noise of the interfering signal is the same when it
tur=
>ns
>> on and off, can I turn off the adaptation when the switch is off, and
tur=
>n
>> on the adaptation when the switch turns on so that my filter doesn't
>> misalign and continue to find the MMSE.
>>
>> Thanks
>
>As Fred stated, the LMS-based adaptive filter works on signals that
>can be correlated. Without knowing the situation you have, I would
>suggest you look at sub-band decomposition adaptive-gain filters.
>Decompose the signal into sub-bands, determine the SNR for each sub-
>band, apply a gain proportional to the SNR for each sub-band, then
>reconstruct by adding the sub-bands. This is a non-parametric method
>for noise cancellation.
>
>Maurice Givens
>

On Jan 21, 9:38=A0am, "pacman101" <janpac01@n_o_s_p_a_m.yahoo.com>
wrote:
> Hi,
>
> That sounds interesting: sub-band decomposition adaptive-gain filters.
>
> I have been doing some searching but I can only find the application to
> equalizers. =A0Could you point a reference where it is applied to noise
> cancellation? =A0That looks really interesting.
>
> I am still trying to quantify the problem. =A0FYI I am not trying to canc=
el
> Acoustic Noise, but RF noise. =A0The noise being generated is right on to=
p of
> my frequency band. =A0The switch is part of a system that for some reason
> generates RF noise (EMI) when it is on but not when it is off. =A0
>
> My receive antenna receives the random noise and that's where I came to
> this idea. =A0I could move the antenna, but I can only move it so far tha=
t I
> still receive the noise. =A0The receive signal is already really weak and
> there is a big difference between the two (about 20-40 dB at different
> bands). =A0 =A0
>
>
>
>
>
> >On Jan 19, 2:23=3DA0pm, "pacman101" <janpac01@n_o_s_p_a_m.yahoo.com>
> >wrote:
> >> Hello,
>
> >> I encountered a problem with Noise Cancellation. =3DA0
>
> >> The interfering signal that I am trying to cancel goes on and off
> really
> >> fast (noise of a switch). =3DA0I have never quantified how long it tur=
ns
> on=3D
> > and
> >> off but it it is less than 1 ms or so.
>
> >> I am attempting to cancel its noise by applying an adaptive filter and
> my
> >> application works for stationary signals such as band-limited white
> >> Gaussian noise, but it doesn't work for this problem.
>
> >> Assuming that the noise of the interfering signal is the same when it
> tur=3D
> >ns
> >> on and off, can I turn off the adaptation when the switch is off, and
> tur=3D
> >n
> >> on the adaptation when the switch turns on so that my filter doesn't
> >> misalign and continue to find the MMSE.
>
> >> Thanks
>
> >As Fred stated, the LMS-based adaptive filter works on signals that
> >can be correlated. Without knowing the situation you have, I would
> >suggest you look at sub-band decomposition adaptive-gain filters.
> >Decompose the signal into sub-bands, determine the SNR for each sub-
> >band, apply a gain proportional to the SNR for each sub-band, then
> >reconstruct by adding the sub-bands. This is a non-parametric method
> >for noise cancellation.
>
> >Maurice Givens- Hide quoted text -
>
> - Show quoted text -

Look up the 1994 first annual comp.dsp conference papers (the second
annual conference was in 2009. Use google and look for comp.dsp
conference). There you will find a paper on noise reduction
techniques, and several sub-band decomposition techniques are
presented.


Maurice Givens

On Jan 24, 12:53&#4294967295;pm, maury <maury...@core.com> wrote:
> On Jan 21, 9:38&#4294967295;am, "pacman101" <janpac01@n_o_s_p_a_m.yahoo.com>
> wrote:
>
>
>
>
>
> > Hi,
>
> > That sounds interesting: sub-band decomposition adaptive-gain filters.
>
> > I have been doing some searching but I can only find the application to
> > equalizers. &#4294967295;Could you point a reference where it is applied to noise
> > cancellation? &#4294967295;That looks really interesting.
>
> > I am still trying to quantify the problem. &#4294967295;FYI I am not trying to cancel
> > Acoustic Noise, but RF noise. &#4294967295;The noise being generated is right on top of
> > my frequency band. &#4294967295;The switch is part of a system that for some reason
> > generates RF noise (EMI) when it is on but not when it is off. &#4294967295;
>
> > My receive antenna receives the random noise and that's where I came to
> > this idea. &#4294967295;I could move the antenna, but I can only move it so far that I
> > still receive the noise. &#4294967295;The receive signal is already really weak and
> > there is a big difference between the two (about 20-40 dB at different
> > bands). &#4294967295; &#4294967295;
>
> > >On Jan 19, 2:23=A0pm, "pacman101" <janpac01@n_o_s_p_a_m.yahoo.com>
> > >wrote:
> > >> Hello,
>
> > >> I encountered a problem with Noise Cancellation. =A0
>
> > >> The interfering signal that I am trying to cancel goes on and off
> > really
> > >> fast (noise of a switch). =A0I have never quantified how long it turns
> > on=
> > > and
> > >> off but it it is less than 1 ms or so.
>
> > >> I am attempting to cancel its noise by applying an adaptive filter and
> > my
> > >> application works for stationary signals such as band-limited white
> > >> Gaussian noise, but it doesn't work for this problem.
>
> > >> Assuming that the noise of the interfering signal is the same when it
> > tur=
> > >ns
> > >> on and off, can I turn off the adaptation when the switch is off, and
> > tur=
> > >n
> > >> on the adaptation when the switch turns on so that my filter doesn't
> > >> misalign and continue to find the MMSE.
>
> > >> Thanks
>
> > >As Fred stated, the LMS-based adaptive filter works on signals that
> > >can be correlated. Without knowing the situation you have, I would
> > >suggest you look at sub-band decomposition adaptive-gain filters.
> > >Decompose the signal into sub-bands, determine the SNR for each sub-
> > >band, apply a gain proportional to the SNR for each sub-band, then
> > >reconstruct by adding the sub-bands. This is a non-parametric method
> > >for noise cancellation.
>
> > >Maurice Givens- Hide quoted text -
>
> > - Show quoted text -
>
> Look up the 1994 first annual comp.dsp conference papers (the second
> annual conference was in 2009. Use google and look for comp.dsp
> conference). There you will find a paper on noise reduction
> techniques, and several sub-band decomposition techniques are
> presented.
>
> Maurice Givens- Hide quoted text -
>
> - Show quoted text -

By the way, it will be called sub-band spectral subtraction