NI USB-6211 noise issue related to sampling frequency

[IDE_R]

Hello,

I am using NI USB 6211 single-ended analog input  for data acquisition.

The problem what I am facing is the noise which has a clear dependency on sample frequency. The noise peak lies generally in between 20 to 30 Hz and in amplitude around 30 dB higher from the base spectrum line.

The interesting observation is that this noise peak dispappers when a sampling frequency like 3kHz, 6 KHz or 9 KHz etc are implemented for the acquisition. Sampling frequency like 2KHz, 4KHz, 8 KHz produce that noise. 

That means if the clock frequency (20 MHz) is  divisible by sampling frequency then that specific noise has been formed.

A figure has been attached herewith. A signal from an input channel has been shown here. The channel has been closed without any incoming signal. It is only noise. In this figure two measurements have been shown. The green curve is the measurement with 3 KHz sampling frequency and the blue curve is the measurement with 4 KHz sampling frequency. At 24 Hz a sharp peak is observable in the blue curve (4 KHz sampling frequency).

I look forward hearing from you on this issue. Thank you in advance.

[Henrik_Volkers]

Blind shot:

Alias (subsampling/mixing) products due to a missing alias filter?

What do you get if you sample with 250kSPS??

Could be a LF radio station or a bad LED light or the interal DC/DC converter.

[johnsold]

Like Henrik, I suspect you are aliasing something. The 3 kHz trace appears to have a peak near 1000.  Suppose there is a signal at 3976 Hz. When sampled at 4 kHz, the aliased signal appears at 24 Hz. When sampled at 3 kHz, the aliased signal appears at 976 Hz.  The actual signal is probably at a higher frequency because you see a similar effect at other sampling frequencies. A careful examination of all the spectra might give you a clue to the actual frequency.

Or, again, as Henrik suggested, sample at a much higher rate and see what is happening.

Lynn

[IDE_R]

Hi Henrik and Lynn,

Thanks for your prompt reply.

Here I have attached two figures.

In one you will find two measurements with 4000 Hz and 4001 Hz sampling rate. Please observe the noise peak differecne coming from this sampling frequency variation (4000 Hz vs 4001 HZ).

In another figure please find the measurement with 125 KHZ sampling.

Thanks in advance.

KI

[johnsold]

The peak at 50 kHz looks very suspicious. It may be from switching power supply noise or something in your system running at that frequency. The amplitudes of the three peaks are quite close.  Notice also that in the 4000/4001 sampling data the two smaller peaks associated with the large peaks ppear to be harmonically related: 26, 52, 78 in green and 134, 268, and 402 in blue.  The high frequency signal has peaks at 500 and 600 kHz and possibly at 700 kHz. 

It is also apparent in the high frequency signal that you have a probable power line frequency signal with harmonics to at least the tenth. (50, 100, 150, 200, 250, 300, 350, 400, 450, 500 Hz). The levels are at or below the baseline on the 4000/4001 spectra, so those peaks are not obvious.

Lynn

[Henrik_Volkers]

Well I don't know the details of your software...

But how do you sample at 4000Hz and 4001Hz ? Do you use external triggering?

If you use the DAQ driver , I have the impression that you just entered the sample rate, hoping the the card is able to sample at these frequencies 😉

However, (without digging into the specs) I don't think that this 'simple' device  is supporting this arbitrary samplerates, and the driver will coerce to the next higher possible samplerate. If  you don't check the property 'actual samplerate'  after configuration, you maybe just misinterpret you data.  

(26 and 134 could be just numbers from the FFT bins and are about factor 5, which might be the next possible samplerate... 

)

OK, again just some shots in the dark 😉

[johnsold]

Henrik,

Good point.

According to the specifications the USB-6211 has an 80 MHz timebase and a 32-bit counter timer. For fs = 4000 the divider is 20000 and the frequency is exactly 4000 Hz. With a divider of 19995 the frequency is 4001.00025... Hz.

Spectral leakage must also be considered.  None of the peaks in the image appear to be exactly centered on the (annotated) largest values in the psectrum graphs.

Lynn