Calculating clock drift for the NI 9234 module

[DennisBengs]

Hello! I'm fairly new to using NI equipment (and hardware DAQ in general), so this is probably going to be a really basic question.

We're planning to use two cDAQ 9181 chassis with two NI 9234 modules to sample data on 8 physical channels simultaneously at a rate of 25600Hz. Since we're using the 9181 chassis we cannot do hardware synchronization as mentioned in this topic:

https://forums.ni.com/t5/Multifunction-DAQ/Trigger-aquisition-across-multiple-compactDAQ-chassis/td-...

"The cDAQ-9181 will be out of the question because, as you stated, it does not provide externally accessible PFI lines and you can't use a digital module for sync in a single module carrier, because that would occupy the only available slot."

We need the synchronization between the channels to be accurate to ±100ms. According to the specifications at http://www.ni.com/pdf/manuals/374238c.pdf, the Internal master timebase (fM) has a frequency of 13.1072 MHz, and an Accuracy of ±50 ppm. I'm trying to calculate the maximum amount of clock drift over 1 hour. Does the chosen sampling rate effect the result?

[Henrik_Volkers]

Well one hour is 3600s, +-50ppm clock uncertainty

worst case 100ppm difference in the clocks  : 3600*100e-6=360ms drift

Say 50ppm is twice the standard deviation (about 95%) and both units are randomly drifting: 3600*sqrt(25^2+25^2)*10-6~ 127ms drift ... almost...

Or share one signal channel and use it to compensate the drift....

Never tried it, but in theory you can even run one IEPE sensor on two current sourcing IEPE inputs. (Well, the sensor usually can handle 2mA to >8mA current...)  

[DennisBengs]

Hi Henrik. Thanks for your reply.

Ah, now I understand how PPM is calculated. I was way off. That clock drift is really quite bad at its worst.

You said you could share a signal. Do you mean like connecting the two NI 9234 modules together using one channel each?

[johnsold · Accepted solution]

I think what Henrik meant was to connect the same signal to one channel on each of the two modules. That way you have two digitized versions of the same signal with each being digitized by a different clock. Provided there are features in the signal which can be identified within the worst case timing difference, you can determine that sample [i] in one data set occurred at the same time (within one sample period) as sample [k] of the other data set.  Then you apply that timing relationship to all the other channels in both modules.

Strictly speaking this refers to the differences in frequency between the two clocks, not drift.  The data sheet specifies the freqeuncy accuracy as +/-50 ppm but does not specify drift at all. Drift usually has two components - temperature and time. Temperature drift is specified as ppm/degree while the time drift (also called aging) may be specified as ppm per year. Both drift coefficients are usually smaller than the accuacy specification. It is not unlikely that the accuracy specification may include drift effects.

If the clocks drift significantly over the duration of your run, you may need to make timing corrections multiple places in the data set.  If the two chassis are physically close together, any temperature changes will likely be very similar for the two devices and aging is usually irrelevant on one hour timescales. I suspect that the initial frequency differences are the only things that will be of concern for you. If a few ppm drift matters, allow both devices to warm up for an hour before starting the test. That should be enough to stabilize the temperatures. 

Lynn

[DennisBengs]

Oh. Now I understand. That's a really clever solution. The con is that it uses one extra channel so you're left with a total of 7. And you need the hardware to connect one sensor to two channels.

I didn't know about the difference between drift and accuracy. Thank you for clearing that up. Leaving the device on to allow the temperature to stabilize is also a good suggestion. I've also noticed that the actual signal from the microphones we have changes during the first minutes we leave the device on. Would you have to be actually sampling data while lettting it warm up?

[johnsold]

 Would you have to be actually sampling data while lettting it warm up?

---

Probably not. Most likely the circuits use approximately the same amount of power whether sampling or not. That assumes that the 12.8 MHz (?) master oscillator runs all the time. The dividers for the actual sample clock may not run until sampling begins but I would guess that the oscillator runs whenever power is on.

Lynn

[DennisBengs]

Thank you for all your help. This clears up a lot of things.