4472 usage below 1 kS/sec

[Michele IFN]

When I use
DAQ_Set_Clock(iDevice, 0, dSampRate, 0, &dActualRate);
to set samplig rate for 4472, dSampRate values below 1 kS/sec are accepted and sent back in dActualRate.
Ni Document ID 1ESE720O states that below 1 kS/sec "the 4472 will experience decreased accuracy and dynamic range".
My questions are:
1 - Which is the smallest sampling frequency supported by the hardware?
2 - Do the anti alias filters works below 1 kS/sec?
3 - Is it possible to have some indication about the 4472 performances at low sampling rate?

[Jack_A.]

The NI-4472 was not designed for or tested at sampling frequencies of less than 1kS/s. So, we don't have any specs for operating at below that.

What type of application are you working on?
What sampling rate does it require?
Are you working in the frequency domain or time domain?
Do you have an absolute accuracy requirement?

These reason I ask these questions is because in is not common to use the NI-4472 with extremely low sampling rates. Here are some things to think about:

If you are looking for very small frequencies, remember that there is a -3 dB filter cutoff at 3.4Hz on the 4472 (this is reduced to 0.5 Hz on the 4472B). So you won't see any frequencies bleow this point.

If you are looking for absolute accuracy in the time domain, the
4472 isn't the best board to use. A 16-bit E-Series has better absolute accuracy specs.

Also, if you have to have a low sampling rate, you can always sample at a faster rate and then decimate the data in software to simulate a slower rate. This way you don't sacrifice any hardware performance.

Hope this helps.
Jack A.

[duncan carter]

I'm in process of porting a commercial application that samples at 64, 128, 256,...,65536 Hz. In non-rigourous tests, my 4474 appears to work well and the anti-aliasing filters appear to work properly. I'll look at this more closely and report back. In the E series version of this application for cases where ther is a minimal hardware anti-aliasing filter, the application samples either at 131072 Hz and then successively applies a .4 sampling rate FIR anti-aliasing filter and then decimates so there would be no problem for me to sample higher on the 4474 and do the same process.

[duncan carter]

Actually, on the E series boards, it samples slightly higher that 131072 Hz and at the end of whatever processing is done I correct for the sampling rate error.

[Michele IFN]

We bought the 4472 to study noise in superconductive electronics (FFT from DC to 50 kHz).

As an additional application we would like to use it for the seismic noise characterization of our lab.

We must acquire 6 seismometer channels at 400 Samples/s and then perform the FFT of the signal. We do not need to cross-correlate the time signal.

The FFT are averaged for many hours. We need a frequency sensitivity better that 1 mHz (in order to fully understand some very narrow mechanical resonances), that could be obtained from time scans of 524288 samples at 400 S/s.

I am aware that these requirements could be satisfied by a time scans of 1048576 samples at 1 kS/s. In fact our software application can reliably manage long samples scans and th
eir FFT, but the memory resources of the system are not enough to perform these operations on 6 channels. The use of double buffering or some other memory saving strategy is possible but would require large modification to the program.

I agree that I could decimate the data and consequently save space on the FFT arrays. But I think that in this case a digital filtering should be used (as suggested in the next answer), with conseguent overhead on memory and CPU.

For this reason I would relax accuracy requirement (as also 16 bits could be enough for the seismometers) and I am investigating the possibility to "undersample" the 4472. As I plan to make relative measurements (comparison between two seismometers), I do not have requirements in linearity. On the contrary I am worry about the proper working of the internal anti-alias filters: we must avoid low frequency ghosts from undersampled lines.

I hope that the -3dB filter cutoff at 3.4 Hz is not active when the ND_DC paramete
r is sent to the 4472.

Thanks
Michele

[Jack_A.]

Yes, the -3dB cutoff frequency is only active when AC coupling is on.

Jack A.

[Jack_A.]

Also, the internal anit-alias filters are never turned off and still work fine, so you don't need to worry about ghosting.

Jack A.