AnsweredAssumed Answered

34465A with DIG option: strange offset in the sample clock rate

Question asked by timop on Apr 2, 2016
Latest reply on Sep 13, 2018 by ealm

I wonder if anybody else has run into this problem, or if an explanation could be found.

When using the 34465A for sampling an input signal (using the DIGITIZE and MEM options), the sample rate is consistently slightly off by about 0.236% from its nominal value. Auto triggering and the fastest timing settings are used, so these should not affect the results.

I first noted this behaviour when digitizing a 1kHz sine signal from a digital audio source, using the highest 50kS/s sample rate. Exporting the sampled signal to Matlab from BenchVue software, an FFT analysis showed that the peak was about 997.65 Hz instead of the nominal 1000 Hz. I verified the frequency of the signal source using the 34465A's frequency measurement function, and this showed a stable frequency of 999.987 Hz, with a deviation span of under 0.002 Hz. 

For some reason, there was an offset of over 2 Hz between the digitized result and the frequency measurement result.

Testing further I connected a GPS receiver 1PPS pulse signal to the DMM and sampled it in a similar fashion. Using the atomic clock signals from satellites, the GPS emits a rising pulse exactly once per second, with timing deviations well below 1 us. Now the DMM sample timing offset was simple to note directly from the time signal: the time difference between adjacent pulses was consistently 50118 samples. So it looks like the sample clock in our 34465A is 118/50000 = 0.236% too fast.

I repeated the same test with a 20kS/s sample rate, with similar results. So it looks like the deviation is not due to errors from dividing a higher frequency sample clock with an inexact integer divisor, but from an offset in the sample clock itself.

Keysight does not describe the accuracy of the ADC sample clock in the technical specifications of the 34465A, nor in the calibration certificate of the unit. However, an error of over 2000 ppm does seem quite large considering the overall accuracy of the DMM model. 

It would be interesting to hear experiences from other users, or any comments from Keysight. If unnoted, this kind of behaviour could lead to unexpected systematic errors using the digitized signals.

Best regards,