Extract high values from graph "PART 2"

[hiNI]

This question builds from the post found here.  

In th attached VI, how to correctly set the threshold so that all the "spikes" can be displayed in the "Spur List"? I should get 3 points in the Spur List.

Thanks,

hiNi 

[hiNI]

Would it be easier if we compare adjacent y-axis values?  If so, if the adjacent y-axis values are beyond a certain threshold, then that will be a "spike"?

[Taki1999]

Try subtracting out a cubic spline fit before you do your thesholding.

Also, your VI has some Rube Goldberg in how you convert your number array to string and then back to number.

[hiNI]

Hello,

I have looked at the examples but not found anything for the Cubic Spline Fit.  Can you assist me in getting your idea to work?

Thanks,

hiNi

[hiNI]

Hello,

I have included the Cubic Spline in the attached VI but not sure how to implement it.  Any help will be greatly appreciated!

Thanks,

hiNi

[Taki1999]

You're nearly there.

Subtract the Cubic Spline Fit from your actual data. You'll be left with a mostly flat dataset with just your spurs sticking out.

With that flattened dataset, you can perform a simple threshold to extract your peaks.

(i.e. don't worry about taking the threshold of your cubic spline fit. That doesn't tell you anything. Do take the threshold of [Your Data Minus Cubic Spline] to get your spurs. I have no idea why you're deleting your first spur.)

EDIT: Also you can expand out your Index Array to pull both the first and second columns. No need to have 2 Index Array functions.

[hiNI]

In this updated data set, is it possible to "smooth out" the data from x = 1000 to x = ~30000 a little more?

The peaks and troughs are a bit much and may skew some of the data catpured from other runs that may result in errenous spike numbers?

[Taki1999]

It's possible.

You'll need to experiment a bit with the fit parameters.

What is the source of your actual data? Do you expect it to follow a theoretical model of some sort? If so, you could fit to the theoretical model and probably end up with a much better initial background fit.

EDIT: You could always brute force it and run it through another pass of background substitution.

[hiNI]

The source of the data is from an insturment that measures phase noise.  The noise plot will vary depending on the circuit being tested.  The data does generally follow the plot like that shown though.

Please clarify your statement of "another pass"...I am not sure what you mean.

Thanks.

[Taki1999]

By "another pass", I mean do another spline fit of your original data - your original spline. Subtract that spline out again to flatten out your data.