How can I detect Turning point in a signal???

[Battousai]

Hi, my problem is detect the turning point of one waveform.... anyone can helpme???

[chilly_charly]

---

Battousai a écrit:   my problem is detect the turning point of one waveform

---

I don't know exactly what you call a "turning point". But I suppose you could calculate the derivative of your signal, then define a threshold and search the derivative for it. There is even a "peak detect.vi" that would easily do the job.

[altenbach]

It would be great if your could attach an image of what you actually mean by "turning point"

We have way too many definitions: http://en.wikipedia.org/wiki/Turning_point! 🙂

[Battousai]

Sorry, ahahaha there are many definitions of turning point. Im looking for the "stationary point" :

For the graph of a one-dimensional function, this corresponds to a point on the graph where the tangent is parallel to the x-axis. For the graph of a two-dimensional function, this corresponds to a point on the graph where the tangent plane is parallel to the xy plane.


About your ideas, when the derivate = 0 the point is a turning point....

How i can derivate the signals??? with the derivate x(i) vi of labview??? its works???


Thanks

[chilly_charly]

---

Battousai wrote: ... How i can derivate the signals??? with the derivate x(i) vi of labview??? its works???

---

Why don't you try it ? What do you fear ? Why would it be so surprizing to find something that works in LabVIEW ? :D:D:D

[altenbach]

A derivative is always problematic if you have noise or if the function is relatively ill behaved. For example if your 2D function looks like an egg carton, you will have many turning points. Do you want the "exact" turning point or is it sufficient to coerce it to an existing x (or x,y) position?

If you have a good mathematical model for the function, I would probably do a nonlinear fit and get the turning point analytically from the best fit paramters. If the function is relatively featureless, a polynomial fit (1D or 2D) might be sufficient for the same.

Do you have some typical sample data? Please post it and maybe I'll try a few things. 🙂

[Battousai]

Im probe with the derivation, and its nots works. You are right, the noise afect the result.

The signals is the sound of one word: Safety.


Thanks

[altenbach]

This sound waveform has a huge number of "turning points". Could you point out which one you actually want?

[Battousai]

I need calculate the number o turning point of the signal to etimate if the signal is stationary or non-stationary.

So, if the number of turning point > 2/3*(N-2)  then the signal is non-stationary
with N= number of points considerate to calculate the turning point

Now, if i choose a window with 400 point, and i move the window until complete the signal, i can make a graphic of the turning point.

Like the example show in the .doc

So, I need calculate the turning point of the signal

I hope you can understandme. (Im spanish, and i dont know write very good in english)


Thanks a lot

[altenbach]

OK, so you want to "count" the (large) number of turning points in a sliding window, getting something related to the average frequency. Tha'ts quite different to the original problem. Maybe there are better algorimthms, how about simply counting the number if sign changes? (or something from the point-by-point palette).