real integration time of HP 34401A

[e61_phil]

Hi,

does anyone know how long the 34401A can really integrate by hardware?

In the service manual of the 3456A I found out that the 3456A is only capable of 10NPLC. The 100 NPLC mode is done by averaging 10 of 10 NPLC measurements.

Is that the same on the 34401A (can't find any hint to it)? In case of only real 10 NPLC in hardware I would use 10 NPLC and SAMPLE:COUNT 10 to get the same result, but better averaging on my computer.

Thanks
Philipp

[Andreas]

Hello,

I think you have also to look after the rounding of the ASCII strings to avoid artefacts.
The output resolution is only ~1uV in 10V range with 10 NPLC
with 100 NPLC it is higher.

with best regards

Andreas

[e61_phil]

The output resolution is only ~1uV in 10V range with 10 NPLC

The question is: Is it real or is it just math (averaging)? I think it is a bit suspicious, that the resolution will drop by one digit at 10V.


I've attached a measurement over many samples (one bin per count) in 100NPLC mode (34401A_8_5digit.png). It looks like wrong binning.


After that I measured the same 10V but with 10NPLC and 10 samples averaged (34401A_10x_10NPLC_1000samples.png).

If the 34401A will just average 10x 10NPLC like the 3456A it seems to be better to do it by my own (on the pc). But I don't want to waste some of the possible rejection if the 34401A is capable of real 100 NPLC. 

[Kleinstein]

If the pre-charge adjustment of the DMM internal AZ circuit is not that perfect, one might be able so see the glitches caused by the AZ switching at the input of the DMM: it might be strong enough to show up on the scope, with just a scope connected to the input in something like 10 V range. I am not sure it is visible, but chances are it can work. This way one could measure the timing used.

The other option would be to carefully look at the noise of 10 PLC and 100 PLC modes. If noise in the 10 PLC mode after averaging 10 samples is lower, it might be worth using the lower noise version. It can be a different result for the 10 V and 100 mV range.

[e61_phil]

If the pre-charge adjustment of the DMM internal AZ circuit is not that perfect, one might be able so see the glitches caused by the AZ switching at the input of the DMM: it might be strong enough to show up on the scope, with just a scope connected to the input in something like 10 V range. I am not sure it is visible, but chances are it can work. This way one could measure the timing used.

The other option would be to carefully look at the noise of 10 PLC and 100 PLC modes. If noise in the 10 PLC mode after averaging 10 samples is lower, it might be worth using the lower noise version. It can be a different result for the 10 V and 100 mV range.

I will do the scope test. But I think it is also possible to do 10x 10NPLC and only one auto zero cycle before (perhaps I should do it this way..).

Is it possible to compare the noise at 0V to get rid of the reference (inside and outside) noise?

[Kleinstein]

For a noise test, one can measure with a short, thus essentially without the noise from the references. One way to look at it would be the Alan variance curve.

Using just one 0 measurement (thus manual "auto" zero) is a two sided way. It increase the noise of the zero measurement (as only a single 10 PLC conversion is used) but would reduce the noise from the readings of the input. In addition it will also hamper the suppression of 1/f noise due to the AZ readings.  So it depends on the signal to measure if this is a good idea. It can help if the measured voltage is very noisy, but it would increase the noise if the measured source is low noise. To a certain degree, spending less time for the zero measurement could be a good idea, but it depends on the signal and a 10:1 ratio would be something for rather noisy signals and it shift the 1/f suppression to around 110 PLC, compared to 20 PLC for for conventional 10 PLC radings.

I have a suspicion the Keithley meters like DMM7510 are doing it a little like this: using less time on zero measurement and use some averaging on the zero reading instead.  At least this could explain there relatively poor noise performance in the 10-100 mHz region, when using the internal AZ mode with a low noise source.

[macboy]

...
I have a suspicion the Keithley meters like DMM7510 are doing it a little like this: using less time on zero measurement and use some averaging on the zero reading instead.  At least this could explain there relatively poor noise performance in the 10-100 mHz region, when using the internal AZ mode with a low noise source.

Quote from Keithley 2001 manual:
• Auto zero off is the fastest.
• Auto zero synchronous is the slowest and most accurate. An auto zero routine is performed for every reading.
• The speed of auto zero normal (default) is between auto zero off and synchronous. An auto zero routine is performed whenever necessary to achieve specified accuracy.

[Cerebus]

If the pre-charge adjustment of the DMM internal AZ circuit is not that perfect, one might be able so see the glitches caused by the AZ switching at the input of the DMM: it might be strong enough to show up on the scope, with just a scope connected to the input in something like 10 V range. I am not sure it is visible, but chances are it can work. This way one could measure the timing used.

The other option would be to carefully look at the noise of 10 PLC and 100 PLC modes. If noise in the 10 PLC mode after averaging 10 samples is lower, it might be worth using the lower noise version. It can be a different result for the 10 V and 100 mV range.

If one is prepared to pop the top off the 34401 a more direct way to assess this would be to stick a scope probe on pin 6 of U106A (an OP-27) - this is the output of the preamp/input to the ADC after the measure/zero/precharge switching and ought to show the action of the former clearly.