Measuring a particular 'deep' in a graph by filtering out the noise

[Intaris]

Altenbach's suggestion is tha way I would go also.

If you have a linear signal with noise, you can fit a line to give more reliable data.

If you have a gaussian peak, you can fit a gaussian function to give you data which will be significantly less susceptible to noise.

There are examples of this shipped with LabVIEW.  Search for "Gaussian" in the NI Example finder.  There's an example of fitting a signal which is the sum of 3 gaussian peaks.

[altenbach]

---

@TITAN-x wrote:

Hello altenbach,

Let me clarify a confusion point. I am not interested in finding peaks. Instead, I would like to measure the "depth" in terms of dB that is shown in the very first image above. The issue I am facing is that; noise level of the signal shown red in color is quite high and I am not able to precisely measure and calculate the depth of the falling part in the signal. Another problem related to the measuring the depth is that; signal keeps swinging up and down. Hence the magnitude keeps changing. What do you suggest me to go with these two issues?

Regards

---

If you fit the peak to the correct function, one of the parameters will be what you are interested in and you can ignore the other fitting parameters.

An recent example for Lorentzian shape can be found here. You have significantly more data over the width of the peak, so the result will be even more reliable. Just optimize the shape of the function to better fit your data. (gaussian, sinc, etc.)

If the width and shape remains constant, another option would be to just integrate the entire peak.

[TITAN-x]

Hello altenbach,

Thank you for your explanatory comment. But I've tried Lorentzian and other functions having similar aims and didn't get what exactly I want. In some of the tried functions I faced with reduction in the deep point (which is the last thing I would accept) that came up while trying to reduce the noise. In some of them, distortions came up and measurement became even worse. I know there is a solution of this problem hidden in somewhere and so that I am keep digging but till now I unfortunately couldn't have a dynamic solution which I can apply to the other graphs having the similar shape (overall shape) but having different parameters (width, deep point and etc.). 

[TITAN-x]

Hello to all LabVIEW experts,

I am doing real time measurement of a particular instrument. Due to some internal disorders out comed graph is not stable. But for a clear measurement procedure graph should be having either no or minimized swinging so that the measurements will be reliable. In this manner, I am in need of your kind aid and suggestions. For a better understanding of the mentioned issue I've attached an example VI that is clearly representing the problem.

P.S 1- We are not allowed to connect any time of the hardware filter to the device.

P.S 2- Pausing the graph & taking measurements is not an acceptable solution as whenever measurement re-started values on the graph is varying due to swinging.

[GerdW]

Hi N.,

But for a clear measurement procedure graph should be having either no or minimized swinging so that the measurements will be reliable.

Are your measurements reliable when the graph does not "swing"? Is this your definition of a good measurement???

For a better understanding of the mentioned issue I've attached an example VI that is clearly representing the problem.

You add some noise to "stable" data: now the plot will "move"…

When the graph should not "swing" you should not add random values!

P.S 1- We are not allowed to connect any time of the hardware filter to the device.

Why this? Hardware filtering most often is much better than software filtering…

P.S 2- Pausing the graph & taking measurements is not an acceptable solution as whenever measurement re-started values on the graph is varying due to swinging.

???

Nobody forces you to pause measurements…

When you get noisy data you need to (try to) eliminate that noise.

Do you know which kind of noise to you expect? Once you are clear about that you can build a filter for…

[GerdW]

Hi N.,

I guess you still want to measure the "deep  in a graph", so I joined your threads.

Why don't you analyze the signal first to determine it's "center" and "baseline" and then plot the filtered signal?

[TITAN-x]

---

Hi GerdW,

Are your measurements reliable when the graph does not "swing"? Is this your definition of a good measurement???

- Yes, exactly. 

You add some noise to "stable" data: now the plot will "move"…

When the graph should not "swing" you should not add random values!

- That was just for you to understand the problem better. I guess it didn't work for you.

Why this? Hardware filtering most often is much better than software filtering…

-As you said most often. Not all the time, that is why we are not allowed.

???

Nobody forces you to pause measurements…

- And no body forces you to write your comments rudely. Why do you do that?

Regards.

[GerdW]

Hi N.,

Pausing … taking measurements is not an acceptable solution

Why is it considered "rude" when I state "you don't need to pause your measurements"?

Did you get my comment on doing the filtering of your signal before displaying it on a graph?

Once you know the center/baseline of that peak you can adjust the signal easily to avoid "swinging of graphs"…