certain frequency artifacts appear in data

[myra]

Hi,

I've noticed in my measurements that there are two particular frequencies that always appear, regardless of what I'm measuring.  I use a DAQ board (PCI 6010) to perform voltage measurements, and Signal Express to control and analyze the data.   I use a differential measurement (between ai0 and ai8, I believe) to measure the voltage output of a piezoelectric (to prevent charging).  I always see a noise signal - artifacts? - at about 4 kHz and at about 9 kHz.  I've found that the input range that I select is extremely important, for instance, if my signal is 20mV p2p and I've chosen an input voltage range of +/- 5 V, there is considerable noise, with these two frequencies very prominent; if I narrow the input range to perhaps +/- 30mV, the noise is much reduced, but the two frequencies are still there.   These signals have appeared in data that I've taken over many months, on various systems, and at different times of data, etc. - but they do not appear if, using the same piezoelectric, I look at the data with an oscilloscope.  

I certainly could simply filter these frequencies out - I use the bandstop filter function to remove the 60 Hz noise, but in my research I'm interested in the frequencies that are excited in a system that are in this frequency range, and I worry that by trying to filter these two out I may also remove some important data along with it.

Is this a common problem?  

Thanks!  Myra

[Corby_B]

Hello myra!

Thanks for your post!

So yes your signal for AI0 should be connected to AI0 for (+) and AI8 for (-). WIth respect to the noise floor, when you set the 6010 in the different ranges it changes the accuracy of the card and so you could notice more noise when setting to a higher range and looking at a smaller signal. See the specs below:

Maximum Voltage Range        -5..5 V
      Range Accuracy                5080 µV
      Range Sensitivity               48.8 µV
Minimum Voltage Range         -0.2..0.2 V
      Range Accuracy                283 µV
      Range Sensitivity               6.4 µV

However I have not seen a situation where the frequencies you are seeing are always present. You say you always see signals at 4 kHz and 9 kHz? When you talk about taking measurements for months is this always with the same piezoelectric sensor type? If you put a signal into your 6010 with a function generator at a known frequency do you still see the 4 and 9 K signals? Let me know if any of this information helps you and let us know your findings!

Cheers!

Corby_B

http://www.ni.com/support 

[myra]

Hi,

Thanks for your reply, and sorry to take so long to respond.  I tried using a function generator (at 10kHz) as input for the DAQ board, and found that the 4kHz signal does appear there.  I was mistaken about the 9kHz signal - it's actually 8kHz, and seems to be a harmonic of the 4kHz.  Depending on how small the signal is, more harmonics seem to appear.  I turned the amplitude of the function generator all the way down and chose an input range of +/- 1V; for this case I could see both the 4 and 8k signal, but they were very small.  When I used the same input amplitude with a range of +/- 5V, the 4 and 8k signal were much larger - and you could faintly see more harmonics in the power spectrum. As before, the same function generator input analyzed by Fourier analysis with my oscilloscope did not reveal those same noise frequencies.

I generally use two similar piezoelectrics (from the same batch) for my measurements, and I've seen this occur with both of them.

I suppose at this point that I've verified that those signals are artificial, so I could just filter them out with the bandstop function - but as I mentioned before, I'm looking at the frequency response of my devices, so would prefer not to remove any important data if at all possible.  

Any suggestions?

Thanks, Myra

[ChrisR.]

Hi, Myra.  I have a couple of questions to help get to the bottom of this.  I don't see any 4kHz tones with a PCI-6010 in my system.

1.  Have you tried looking at the internal ground channel.  i.e. _aignd_vs_aignd?  That will help determine whether the noise is being picked up externally or is internal to the board itself.

2.  Have you tried different sample rates?  It's possible that the noise is at a different frequency but is aliasing to 4kHz.  Try moving your sample rate a little bit and see if the tones change their apparent frequency.

Chris 

[myra]

Hi,

Thanks for your suggestions. 

I'm not sure how to look at the internal ground channel - do you mean I should set up a step (I'm using Signal Express) to look at the channel that corresponds to ground?  I need to pull up the wiring diagram because I can't remember what all the input names correspond to.

I tried changing the sampling rate as you suggested - I generally use 200kHz, although occasionally have used 100kHz.  Just now I tried sampling rates of 200, 187, 150, 100, and 20kHz; the noise signal was always near 4kHz but did vary very slightly - from 4.045 to 4.08k at the most.  More precisely, the positions were:

sampling rate (kHz)      noise freq (kHz)

      200                                4.05

      187                                4.052

      150                                4.08

      100                                4.075

       20                                 4.045

I don't see a pattern there...  I thought aliasing is mainly an issue when your sampling rate is too low for the frequency being measured?  The frequencies I look at are generally between 6 and 22 kHz, and I generally sample at 200kHz (the max for my card).  I know that the minimum necessary is 2x, but 10x is better!  Are there other instances of aliasing that you refer to? 

Thanks, Myra

[ChrisR.]

I have no experience with Signal Express, so I can't help you with that.  I was just suggesting that instead of measuring the output of your sensors, you could try measuring ground, preferably internally, just to eliminate one possible source of the noise.

It looks like the noise really is at 4 kHz.  What I was looking for was whether it could have been something like 204kHz.  204kHz noise sampled at 200kS/s is indistinguishable from 4kHz.  The same would be true if it were sampled at 100kS/s.  I was thinking you'd try smaller variations in sample rate.  For example, if you changed your sample rate to 198kS/s, then 204kHz noise would appear in the spectrum at 6kHz, whereas 4kHz noise would still be at 4kHz.

Have you tried the board in a different computer?  It's possible the computer is the source of the noise.

Chris 

[Corby_B]

Hello myra,

I agree with Chris in that your noise seems to be a constant value of about 4KHz. This would indicate to me that it is more than likely an external source that is causing the noise. If you try a grounded measurement (RSE) so connection your incoming voltage signal to AI0 (pin 68) and AI ground (pin 67) to see if that eliminates or reduces any of the noise.  Let us know your findings!

Cheers!

Corby_B

http://www.ni.com/support 

[myra]

Hi,

Thanks for your input!

I've wondered if the noise originated from the computer - what else would create such a high frequency?  I thought that I had disproved it by showing that the same measurement (with the piezoelectric physically in the same spot, by the computer) made with the oscilloscope did not contain the 4k signal; however, I hadn't thought about how the noise could affect the DAQ connector board.  I'm using the CB-37F-LP connector block, which comes unshielded.  I made a little steel box for it, with rubber offsets and feet.  Maybe I need more shielding though?  What would you recommend?

I've made grounded measurements before, but since a piezoelectric does not have its own ground it became charged to the lower or upper register almost immediately - that's why I switched to the differential measurement.  I thought that the differential measurement should be better in general anyways, because of common mode rejection.  What does that imply about the noise that I'm seeing?  How would it get through on only one line?  I believe that I've balanced the receiving end, but is it possible that the rest of my measurement is not balanced?  I'm not sure what else I would need to do. 

Is there a way in signal express to look at the ground channels?  I can't choose them directly when setting up a step, but perhaps there is another way?  Or maybe there is a Labview vi that I could use?  Or something in MAX?  I played around with it yesterday but to no avail. 

Thanks, Myra

[myra]

Sorry - I forgot to add that I was wondering why, if the problem is a noise source at 4kHz, I tend to see so many harmonics too?  The 8kHz one is always there, and generally many higher ones.  That is why I thought that it might have something to do with the DAQ board itself (or the connector block).  If it were aliasing from a much higher frequency, above my 200kHz sampling rate, would these noise frequencies appear this way?  I thought that the image frequencies generated are abs(freq - N*sampling freq), so with my high sampling rate the image frequencies would be spaced by 200kHz, right?  So it must be something else....

Myra

[Corby_B]

Hello myra!

Thanks for your post back!  I think we might have to bring this down to the basics. If you have the ability, can you put a known frequency signal in an AI port of the 6010 by way of a function generator? For instance a 1KHz sine wave? If you see these other frequency components here then that could indicate a problem with the DAQ board or the computer that it is in. The fact that you are using differential is great for what you are doing but sometimes you can have a situation where access current can build up. For instance when taking thermocouple measurements you sometimes need a ground reference for current to travel or else it will educe noise to your voltage reading. Take a look at Figure 1 in the following link. 

Field Wiring and Noise Considerations for Analog Signals

 Also if your sampling rate is 200K and you were sending frequencies much higher than that then yes you could also see the harmonics of that signal. Not sure how the DAQ board would resolve those frequencies however but it would be an order of the frequency being sampled and not based on the sample rate. I would be interested to see if you still have these components if you sample a known signal. You could also do a "loop back" best on your DAQ board. Generate a known signal on an AO channel and feed it directly back to an AI channel. Let us know if any of this information helps you with the frequencies that you see!

Cheers!

Corby_B