Which noise?

Dear chaps,

I have a problem. I have a signal recorded from a ultrasonic
transducer of a chemical reaction. It is rich of noise, expecially the
white one. I tried to filter with normal filter (butterworth band
pass), but I did not obtained a lot of success.
So, I thought was better delete the White gaussian noise with an
adaptive filter. I used the demo of Simuling of LMS but I have
problems to "create" the noise artificially?
What can I do? Which information I need to identify it?
Is it enought to record the noise from the becker without reaction?

Does it work or I must think about a different filter?

Thanks in advance!
Manuel

Manuel Tramontana wrote:

> Dear chaps,
> 
> I have a problem. I have a signal recorded from a ultrasonic
> transducer of a chemical reaction. [snip]

Any links to an overview of the setup you are using?

Based on experiences I had 30 years ago as an electronics tech for a 
university chemistry department, I suspect it will very helpful.

"Manuel Tramontana" <tramoman@yahoo.it> wrote in message
news:38b6dcc4.0402020953.41785fec@posting.google.com...
> Dear chaps,
>
> I have a problem. I have a signal recorded from a ultrasonic
> transducer of a chemical reaction. It is rich of noise, expecially the
> white one. I tried to filter with normal filter (butterworth band
> pass), but I did not obtained a lot of success.
> So, I thought was better delete the White gaussian noise with an
> adaptive filter. I used the demo of Simuling of LMS but I have
> problems to "create" the noise artificially?
> What can I do? Which information I need to identify it?
> Is it enought to record the noise from the becker without reaction?
>
> Does it work or I must think about a different filter?

There are some things you haven't described that would be important:

If you used a bandpass filter, how did you select the band center and band
width?  There must be something about the signal that is known???

How do you know what part of the signal is contaminating noise and how much
of the noiselike character is the signal of interest?

Is there anything you might do better with the sensor to eliminate noise?

As far as adaptive filtering is concerned, you might try an enhancer
version:

Input and filter are in parallel and the filter output adds (or subtracts)
with the "direct path" to form a "system output".
Adapt the filter to reduce the system output.
Because the "interference" is noise, the best the filter can do is to shut
off at the noise frequencies and to cancel the signal at the signal
frequencies.
That's why this rather direct method won't work.
However, if you look at the output of the filter rather than the output of
the "system" described above, all frequencies that contain only noise are
attenuated.
If the signal of interest has spectral lines, the latter will possibly work.
If not, then it likely won't.

         Please view in a fixed-width font such as Courier.



                                   +--------------------+
                                   |                    |
                                   |                    |
                                   |                    |
  input------+---------------------|------------>(+)----+----> e[n] "system
output"
             |                     |              ^
             |                     |              |
             |                     v              |
             +------------------>[LMS]------------+----------> o[n] "filter
output"


    Adaptive enhancer:
    LMS adaptive filter adjusts to minimize e[n] which cancels signal
    and adds no additional noise.
    o[n] ideally contains only signal


Fred