Data precision in spectral measurement

[ZeSuli]

Hi guys,

I'm using DAQ assistant and Spectral Measurements to analyze a signal. I need the exact distribution between 49.8-50.2Hz but the both datas acquired from the "Spectral Mesurement" and "FFT Powerspectrum and PSD" seems to have a too rough datas and only power of the frequencies of integer is showed. Is there any way I can acquire more precise datas? I've heard of NI-RFmx SpecAn, will that be helpful?

Best Regards

[Bob_Schor]

Something is definitely strange.  I wonder what you know about Signal Analysis, and the relationships among sampling frequency, sampling time (or number of samples), the resulting frequency range and the frequency resolution (the smallest frequency different in the spectrum).

Your DAQ Assistant (shudder!!) suggests you are sampling 25,000 points per second, for a second.  How did you get this frequency plot?  Are you really sampling signals over a 2kV range?  

Why (oh, why?) is there a Timed Loop here?  Doesn't your DAQ device do the timing for you?  [Oh, you're using the Dreaded DAQ Assistant, so maybe you don't understand Hardware Clocks in DAQ devices ...].

Bob Schor

[ZeSuli]

Hi Bob_Schor,

Thank you for your reply. I'm a beginner and really have no idea where to begin with. So where is there any way to solve this, or at least, make it better? A tutorial or whatever recommendations are appreciated.

Best Regards

[Bob_Schor]

Sure.  Look up Frequency Analysis, Spectrum, FFT, similar topics.  Actually, those "key words" might lead you astray -- try Signal Processing (which goes to Wikipedia).  Anyway, it doesn't make sense to do "sampling" and "frequency analysis" if you don't understand what it means to sample a signal, nor what "frequency" means.  Do you know what FFT means?  Can you define/explain it?  For extra credit, why did the first "F" get added, and why is it important?  And who are Cooley and Tukey, anyway?

Bob Schor

[Kevin_Price · Accepted solution]

A short (and partial) answer is that to get better frequency resolution, you need to collect data for a longer time.  There's a simple relationship:  df = 1/T  where df is the delta frequency between points in your spectral plot and T is the total capture time of the original data fed into the spectral analysis.

In short, to get 0.1 Hz resolution, you need to capture and analyze 10 sec chunks of data.  For 0.01 Hz resolution, you need 100 sec chunks.   Etc.

More generally, I'd like to offer this up as a good beginner's lesson.  Before you start believing and trusting the results of your analysis, be sure you understand the basic principles of that analysis first.  Here, df = 1/T is one of those basic principles.

Not a criticism, just a friendly caution.  There's a lot of signal processing functions that you *can* use in LabVIEW, but it'll always be important know which ones you *should* use (and how to use them) before you interpret meaning out of their results. 

Actually, you already did pretty well here because you questioned your results.  You noticed that the frequency resolution wasn't as fine as you needed it to be, and recognized that you'd better figure out how to improve that.  As a present-day beginner, keep that kind of mindset going forward.  

-Kevin P

[afilipek]

It may also be helpful to take a look at this public document, starting from page 15: http://www.meo.etc.upt.ro/materii/cursuri/PG_MASTER/LVSP.pdf 

Personally, I would start from the topic:

- how the ADC works

- sampling quantization coding

- spectrum analysis

- drop on a block diagram the Simulate Signal, FFT and Waveform Graph and play with it

Best,

Andrzej

[Bob_Schor]

---

@afilipek wrote:

It may also be helpful to take a look at this public document, starting from page 15: http://www.meo.etc.upt.ro/materii/cursuri/PG_MASTER/LVSP.pdf 

If you have difficulty accessing the LabVIEW Signal Processing Course Manual, Version 1.0 (September 1997) from the URL that Andrzej posted, just try a Google search for LVSP.pdf, which worked for me.  I should note that this uses LabVIEW 4.0, so some of the Block Diagrams may look a little strange ...

Doing a little wider searching, you can find other reference works on digital signal processing, including some that use LabVIEW.  If you have access to a good library (say at a School of Engineering), there should be plenty of material there, as well.

Bob Schor