Fitting complicated step functions

[emrahturgut]

Hi

I have complicated step function as in image. I need to get the values in each step, what is the fastest fitting procedure?

[GerdW]

Hi cornell,

usually one tries to detect the steps in your signal and average the parts in between…

What have you tried so far?

[emrahturgut]

Thank you for the reply. I am trying to fit it, find the derivatives and thresholds etc. Then average the values between indeces. But it starts to get complicated and I want to ask if there is something build in. 

[GerdW]

Hi cornell,

it starts to get complicated

You think to analyze your signal as shown is complicated? With easily detectable steps and nearly no noise?

I want to ask if there is something build in

Did you check the signal processing palette?

General hints:

- getting the derivate is to subtract two samples and divide by the time between them…

- your signal seems to have steps in a certain frequency so it might help to discover them as you already know when to expect the next step…

- to detect a step you could check if the current value is outside a range around a running average of the last x samples (range should be greater than expected noise level)…

[emrahturgut]

I agree it is a clear signal, but that fitting has to to happen for each pixel of an image. And I am using DAQ with 30kHz rate. I am looking for a tricky way of doing it. 

[crossrulz]

---

@emrahturgut wrote:

And I am using DAQ with 30kHz rate. I am looking for a tricky way of doing it.

---

Let's explore this a little more.  Where is the signal coming from?  Does the source of the signal have a sample clock?  What is the rate of that clock?  Is is exportable so that you can sample it yourself (or use it as a sample clock yourself)?

[GerdW]

Hi,

You don't want to fit, you want to split and average.
Have you tried my ideas/approaches?

Attach a VI with some default data in it!

[mcduff]

---

@emrahturgut wrote:

I am looking for a tricky way of doing it. 

---

If you are familar with wavelets then read on. Your signal looks taylor made for Haar wavelets. Unfortunately they are not in the base package, but you could program them in if necessary. Look at tutorials for wavelets if you are unfamiliar.

cheers,

mcduff

[emrahturgut]

Here are some data. 

[emrahturgut]

There are four signal read by DAQ. 

Two has 3kHz modulation, two is AI voltage. 

It needs to take 1-D scan at 100 pixel. But in each pixel it needs to sample ~200 points, so that it can sense the 3kHz modulation (DAQ rate is in order of 20kHz ). 

For each pixel, it needs to find the average of two AI voltage as shown in the figure. 

and find the peak to peak values in the modulated signal. 

In addition, it does the 1-D scan ~1000 times. 

I am planning increase 3kHz modulation and number of pixel later. That is why I am looking for some elegant way of doing it. 

Bests