FFT analysis with 24bit soundcard vs. 18bit NI-DAQ ?

[Gustep]

Hello,

I am currently using an 18-bit NI-DAQ ADC / DAC card (PCI-6281) to
perform some sensitive tone level and distortion measurements over time
via FFTs. As a preamp, I am using the LT1678, which has exceptional
distortion and noise specs, particular at low frequencies - the 1/f
corner frequency is 5Hz or so. I also tried the same measurement with a
comparable 16-bit NI-DAQ card (PCI-6221), and found that the noise in
my measurements did increase.

So now that it appears that bit resolution matters in my measurements,
I am thinking about 24bit acquisition. National Instruments has some
very expensive 24-bit DSA cards like the PCI-4461, which I'd love to try, but these are too expensive to justify the purchase if I just want to try it. (I wonder if I could rent one or test it for a week...?). But maybe one of the many 24bit soundcards could also help me gauge the benefit of 24 bits. Especially the EMU 1212M is affordable and has a low noise
floor. The M-Audio Delta 1010LT is almost as good, and has 8 ADC inputs and 8 DAC outputs, so that's a great candidate, too.

So my question is, has someone already tried and compared NI-DAQ cards
to good sound cards for tone analysis? E.g., measuring the amplitude
stability of a 1kHz tone over time?

I did read a related post <a href="http://www.diyaudio.com/forums/showthread.php?postid=690127#post690127" target="_blank" rel="nofollow">here</a>,
which claims that even a 16bit NI-DAQ card is better than a 24bit sound
card, but a few questions remain. For example, what VI was used for this
comparison? I think LabView default sound VIs are limited to only 16bits when acquiring data from sound cards, unfortunately (this may have changed with more recent versions of Labview). But I also know that you
can get some improved VIs over the web, for example <a href="http://www.zeitnitz.de/Christian/WaveIO/waveio.html" target="_blank" rel="nofollow">here</a>, which allow you to acquire up to 32bit into LabView 7.1 and up over your soundcard!

So does someone have experience with using soundcards with 24bit VIs and labview to do spectral analysis work? Any tips and sharing of test results is greatly appreciated.

[Joe F.]

Hi Gustep,The increase in resolution from 16- to 18-bit M Series cards will definitely improve the measurements you make. Better yet, switching to a proper DSA board will give you the best results. The M Series design is multi-purpose, so it tries to balance DC accuracy and AC clarity, but a DSA board is tailored for AC analysis and will give you the best dynamic range. The best way to see if you can test-drive one of these cards is by contacting a technical sales representative at [phone].I haven't seen any official comparisons done between NI DAQ cards and the various sound cards. Perhaps someone in the community has done a test or evaluation.

[Chris R.]

Which product works best is very much a function of what you're trying to do. You mentioned measuring the amplitude stability of a 1kHz tone over time. The 628x DAQ product will do much better than any DSA product for this, as the DSA products are not designed for good gain and offset stability. The 407x DMM products will do even better, by the way, for this particular application. I would not expect good gain stability from any kind of sound card. My understanding is that you care about amplitude stability and distortion. It's important to know what frequencies you're interested in and how good the stability and distortion need to be. The 628x boards have good distortion at low frequencies. At very low frequencies, the 407x DMMs will give distortion around the 1ppm (-120 dBc) level.If you contact a sales rep to get a loaner PCI-4461 to evaluate, see if he/she will loan you a PCI-4070 as well, unless you know that one channel won't be enough. Also, have you considered the PCI-4474, a DSA product that costs less than the PCI-4461?Chris

[Gustep]

Thanks to everyone for the helpful answers. Yes, my principal
measurement consists of precisely measuring the amplitude stability of
several tones on one line. They are in the 200Hz to 5kHz range. Since I
have several tones on the same signal line, which I need to measure
independently, the DMM series may not work despite the better SNR.

[Chris R.]

Hi, Gustep. The 407x DMMs have a digitizer mode that preserves the low-frequency performance of the DMM. The signal path is the same for digitizer mode and DMM mode. You can do FFTs and separate the tones just like you can with a DAQ board. Can I ask what range you're working with? +/-10V? What's your signal amplitude? And what degree of amplitude stability are you trying to measure? That is, +/- how many parts per million? You should be able to get really good results with both the 628x and the 407x products.Chris

[Gustep]

Hi Chris,I have a +/- 4.5V (9 Vpp) signal consisting of several tones, and I'm trying to resolve the tone amplitudes to within 3 microvolts (rms) or better if possible. That's roughly 1 ppm if I am not mistaken. I manage to get approximately this performance out of the PCI-6281 with a 1-second, 10kHz sample rate FFT, which results in a noise floor of roughly 2.5 microvolts (rms). For a single-tone signal at max amplitude, this noise floor would be around -120dBc, I think. To get this performance in 8-channel mode, I had to put eight analog anti-alias filters (-3 dB at ca. 10kHz) between my device and the signal inputs of the PCI-6281, one in each signal line. And furthermore, which components I use in my analog frontend also makes a difference.If possible, I would like to improve the measurement resolution further, and two things caught my attention. First, switching over to a PCI-6221 decreased my measurement precision, all other things being equal. Second, the sales literature of the 24-bit DSA boards featured a beautiful FFT with a -150dBc noise floor on a single-tone signal. But thanks to your info on the digitizer mode of the 407x, I'll also look into that, even though apparently they only have up to 4 channels.Message Edited by Gustep on 07-14-2008 08:00 PMMessage Edited by Gustep on 07-14-2008 08:01 PMMessage Edited by Gustep on 07-14-2008 08:02 PM

[Chris R.]

Very interesting. At this point I think your best bet is the 6281. Its noise density on the +/-5V range is lower than that of the 4070 on the +/-10V range. The DMM offers ranges only in powers of 10. Although I suppose you could adjust your preamp to utilize the full +/-10V range of the DMM. But anyway, the DMM is a single-channel device, and it sounds like you need more channels. Your sample rate of 10kS/s seems low to me. You'll get a lower noise density (i.e. lower noise for the same measurement time) with as fast a sample rate as you can manage, because that will reduce the aliasing of broadband noise. I'm worried about the multi=channel case, though. You may get "ghosting" and crosstalk, which couples signals from one channel to another and could easily corrupt measurements beyond the 1ppm level. In single-channel mode, you can enable the onboard lowpass filter, which lowers noise density somewhat.I'm curious how you're testing your setup. What AC signal source do you have that is stable to 1ppm? Regarding the DSA boards, I don't know their noise density offhand, nor do I have one to test. I'll try to find out how it compares. Its short-term stability may be adequate, but I'm pretty sure over the long term it will drift around much more than the 6281, which in turn will drift somewhat more than the 4070.Chris

[Gustep]

It turns out that the PCI-6281 analog out is among the best AC signal sources that I have at this point - it is definitely better than a $1500 discrete function generator that I was using until just recently. I still don't know if it is accurate to 1ppm - how is this defined? The amplitude accuracy from one sine cycle to the next sine cycle? I'd say, the 1-second average amplitude of as many sine cycles (i.e. the 500-cycle average amplitude for 500Hz) is probably reproducible to within 1ppm, i.e. from one second to the next. I know that the AO is only 16 bits on this card, but some discrete generators have only 14 bits. Are you suggesting that this is one area that could be improved? I'm already sending the AO signal (DAC sampling rate is 100kHz or so) through a low-pass filter after generation, so I think as long as the 16-bit output is stable, the bit depth at the AO may not be a limiting factor right now.I am indeed worried about settling time, which is why I try to find a compromise between sampling rate and channel cross-talk. Overall, cross-talk is a lesser concern, since the signals on different channels are largely identical and change very little over time, meaning that even with a bit of cross-talk, the final effect may be acceptable. I ran several cross-talk tests, and 10kHz seemed to work well, also in part because the FFT bins fall on nice frequencies this way, but yes, I may try to optimize that again. I don't want to lower the impedance of my AA filter too much, however, otherwise the switching noise from the MUX on the PCI-6281 seems to feed back into my analog frontend and exacerbate the noise.

[Chris R.]

Well if the source of the signals you want to measure is the 6281, then measuring with the 6281 is a good idea, because the analog input and analog output circuits share the same reference on that board. Thus any drift in the reference will be ignored. I agree that it is one of the most stable AC sources around. I once tried to evaluate the measurement noise of the 4070 DMM for AC volts measurements, and I got better results using an M-series analog output, probably on a 6289, than I did with a Fluke 5700A calibrator or any other source I could find. It sounds like you know what you're doing, and are watching out for the right things. I'd be interested in hearing more about your application if you're comfortable posting it.Chris