Measuring Volatge Amplitude using NI-USB 6361

[zakariarahman]

Hello,

I have a NI 6361 DaQ device. I want to measure the rms output voltage of my circuit using one of the analog input channel of the DAq device. The problem is from the specification list I can see the maximum sample rate when using a single channel is 2 MS/s. The frequency of my output voltage is around 7MHz. I want to know since I dont care about the output waveform shape just want to know its accurate rms value will it be possible to measure the rms values using the DaQ device I have. If so what would can i do to make the measurement more accurate. Like take samples for a particular period of time and then find the average rms value?

Thank you very much

[Yddet]

Hello,

It depends but generally no. What accuracy, precision and latency do you accept ?

The more samples you will have the better. But if the input signal is varying there will be a delay in the measure or worse, you won't see the variation. 

I made a small simulator (see attached). Try it with 70Hz input signal and 20Hz sampling rate. You will see the influence of the number of samples.

There is also a risk the samples give a very bad representation of the reality.

If you are very vigilant and never trust the results completely, then maybe it can be used for a test or two. But I would never use this board daily or for a customer.

In my limited experience I have not heard of ways to make the measurement more accurate.

Yddet

[BertMcMahan]

It depends on the bandwidth of your signal. You say it's happening around 7 MHz; what is the lowest frequency component? If it's 0 Hz, then no, your device cannot do this. If it's something like 6 MHz, then... technically, yes it might work.

Look up Nyquist sampling theorem. (Theoretically speaking) to perfectly capture a signal of bandwidth B, you need to sample at 2*B. If your signal ranges from 6-8 MHz, then Nyquist says you must sample at 4 MHz.

The important thing to remember is that when you sample it, you will NOT see the signal accurately reproduced; you will see an alias of this signal. You must properly reconstruct your input signal and apply filters to it to upsample it back into the real world signal. This is not a trivial task, but the math says it can be done. Although Nyquist says 2x frequency I have found in the real world I want to be at least 10x frequency, but you might be able to make it work with some more advanced math.

[Kevin_Price]

While agreeing with all the caveats from Yddet, here are some further thoughts on things to try if you're stuck with the board you have.

1. Realize that you are very much undersampling your signal.  Unless you happen to be sampling at an integer divisor of the signal frequency, you'll tend to be accumulating samples representing many different phase points in your original waveform.

2. If the input waveform can be assumed to be pretty steady, this may not be *all that much* different than the many different phase points you get from 50x oversampling.

    IT IS NOT THE SAME!  But it may be a small enough difference to remain useful.

3. More samples will tend to make your calcs more consistent, and closer to accurate.  (But you cannot count on them being *actually* accurate.)

4. To guard against coincidental cases of sampling at an integer divisor, I'd probably try taking measurements at a few different sample rates.  Again, if the signal is pretty steady and the sample rate picks out points at pseudo-random phases, the RMS measurement shouldn't vary much with different sample rates.  Any outliers can be tossed out.

-Kevin P

[Edit: P.S.   Applications for undersampling are relatively rare, but it's kinda cool when you can use an unconventional technique to get results that might otherwise seem impossible.  My memory was jogged and I just took a quick trip down memory lane back to 2006 for one of them.   For those interested, check out this post (and the rest of the thread it's in).]