sleek/plane a Signal

[CamilloN]

Hello,

i got a problem. I wanna know the dy/dx of a Signal.

The signal is a array from a txt-file.

To calculate the value isn't the Problem, but the Signal is

not clean it has a few peaks from extern interferences.

To calculate the value by hand i normally take two points

on the x-axis and and look at the value of the y-axis...dy/dx

and i got the result.

If i make the width of the two points big, my result is not exact.

If i make it to small the peak-influence is to big.

How can i plane (is that the correct word for my intention???)

the Signal WITHOUT moving the Signal to the left or to the right.

Or is there another Method to calculate the dy/dx

kind regards

Vincenzo Nuzzi

[Tim_S]

QUOTE(CamilloN @ Feb 25 2008, 04:07 AM)

How can i plane (is that the correct word for my intention???)

the Signal WITHOUT moving the Signal to the left or to the right.

Or is there another Method to calculate the dy/dx

I believe the word you're looking for is "filter". Time/phase shifting can be a problem with filters. There are some that are designed to have no phase shifting but those tend to affect amplitude instead. Depending on the filter settings and your data, neither phase nor amplitude may be affected in the areas you want to keep.

You may want to try a "running average" filter. This involves taking multiple sequential points and averaging them; the effect tends to take out little spikes. The number of points you use can impact the portion of your data that you want to keep.

[ASTDan]

I would recomend using a butterworth filter to filter out your unwanted frequences.

Then use the derivitive vi to calculate dx/dy

Dan

[jdunham]

QUOTE(Tim_S @ Feb 25 2008, 03:30 AM)

I believe the word you're looking for is "filter". Time/phase shifting can be a problem with filters. There are some that are designed to have no phase shifting but those tend to affect amplitude instead. Depending on the filter settings and your data, neither phase nor amplitude may be affected in the areas you want to keep.

You may want to try a "running average" filter. This involves taking multiple sequential points and averaging them; the effect tends to take out little spikes. The number of points you use can impact the portion of your data that you want to keep.

The median filter is probably the best at removing spikes. It doesn't add any phase delay. The filter is nonlinear, so it's harder to analyze than a moving average or butterworth filter, but if your only goal is to remove noise spikes which are polluting your data, then it's the best.

[CamilloN]

of course the word was filter :headbang:

With the median and the butterworth filter my results aren't as good as the

"running average" filter.

What if i make a fast fourier transformation than overwrite the unwished

frequencies with zeros and make than a inverse FFT? This could also be

possible, or are there any problems.

Now a question to all mathematician out there ;-)

I know that the FFT mirrors the frequencies. What happens if i also overwrite

the mirror-frequencies? Are they necessary for the inverse transformation?

[Tim_S]

QUOTE(CamilloN @ Feb 28 2008, 02:41 AM)

What if i make a fast fourier transformation than overwrite the unwished

frequencies with zeros and make than a inverse FFT? This could also be

possible, or are there any problems.

Now a question to all mathematician out there ;-)

I know that the FFT mirrors the frequencies. What happens if i also overwrite

the mirror-frequencies? Are they necessary for the inverse transformation?

I've used an inverse-FFT filter before, though it was using the one-sided versions. It worked rather well so long as the frequency content of the noise was "sufficiently far away" from any frequency content of the signal. You will get distortion of the original signal if the signal and noise intermix. With the signals I had, there was a fine line between not eliminating noise and eliminating signal, so you may have to play around with the frequency threshold more than with other methods.