FFT

[erezg]

Hey,

I'm using the FFT.vi and I found that for long signals (>1000000 points) the amplitude I get is smaller then expected (If I break the input to 10000 points and average the output I do get the correct results). Strange......

thanks, Erez

[altenbach]

Do you have some VI to demonstrate? What amplitude did you expect? What amplitude did you get?

[altenbach]

Here's the normalized FFT max magnitude for sizes of 100k ... 2M. They all look the same!

[erezg]

thanks for the fast reply. I used the below (sample rate is 10kKz at 60hz grid):

For N=1000 I get Amp = 0.0707 (as expected)

For N = 10000 Amp =0.69

For N = 100000 Amp =0.66

For N = 1200000 Amp =0.55

For N = 1500000 Amp =0.705

[altenbach]

Please attach your VI instead of an image (we can't even tell what enum you selected!).

What's the name of that subVI? Is that from a toolkit? What does it do?

Did you look at the waveforms before FFT?

[erezg]

yepe, this VI is from the toolkit

thanks, Erez

[GerdW]

Hi Erez,

seems to be some kind of aliasing effect:

Maybe because a FFT prefers blocks with 2^n samples?

[erezg]

Hi,

still strange. If a 1000 points gives good answer, I would expect that any multiplication of 1000 will work well.

In the "real world" starts with sampling of a 60Hz grid with 10kHz sample rate - so every 1000  point is 6 waves (100msec), so any multiplication of 1000 should work.

BTW, according to the FFT help, it does not require 2^N points.

thanks for the help, 

[altenbach]

---

@erezg wrote:

yepe, this VI is from the toolkit

---

What toolkit? What does it actually do? Do you have a link to the online help for it? What happens if you just take the normalized magnitude like I do?

I don't see where you ensure that you have an integer number of periods. Maybe try to compare Ns that are an integer multiple of 167. In any other case, you'll have the FFT peak spread over more than one bin due to spectral leakage.

[erezg]

Hi

this sub vi is calculation the normalized amp:

But I think I found the reason. For long time scan the frequency resolution is better, and I get a wider pick. In order to calculate the real pick we need to integrate the pick.