Electronics > Metrology

3458A autozero on/off for uV measurements

<< < (2/2)

lymex beat me to it.


--- Quote from: zlymex on September 06, 2016, 01:19:20 pm ---File attached. But I'll erase it after 24hrs or someone oppose it.

--- End quote ---

Thanks lymex and to others who offered a copy. Question is did anyone understand it?  :-//

I found it a bit hard going. First observation is that 'is the 3457A's autozero switch really after the input amplifier (IVAAC) as they hypothesise?' I would expect it to be as close to the input as possible, before the amplifier to minimise the amplifier and ADC's low frequency noise and especially flicker noise. However I couldn't see it in the circuit diagram - perhaps it's in the input hybrid U101? As to the 3458A I'm pretty sure the AZ switch is before the DC amplifier (as the service manual says it is before the AC and DC conditioning) but again I can't see it in the circuit.

One of their main issues is the amount of residual DC offset despite the autozero:

--- Quote --- In the process of examining the nonrepeatability, we gradually became aware of two facts:  i) behaviors of the zero voltage error were in a statistical sense almost the same as what we had experienced when we had tried to evaluate the performances of the sensor and of the core noise measurement system developed by us and ii) the statistically averaged value, in spite of increasing the measured values, fell into stagnation without approaching zero, no matter what NPLC was chosen. In relation to this, we tried to examine whether or not the same thing occurs on other measurement ranges. The result was common to every measurement range. The zero voltage error at AZ-ON ?uctuates at least over the last one or two digits, while at AZOFF over the last two or three irrespective of NPLC chosen even for NPLC>1
--- End quote ---

(Really? The last three [digits], of a 6.5 digit meter, were noise with AZ off and NPLC > 1? And that at 0V, so even worse at higher voltages where the reference noise also becomes a factor).

But surely the residual DC offset is to be expected?  Autozero can't eliminate offsets that arise between the input short and the AZ switch including thermal EMF's at the input terminals, relay contacts and other wiring/interconnections/circuitry. There will also be offsets before the AZ switch due to the input leakage currents (amplfier input bias, input protection diodes, input selection fet switches etc.) flowing to ground through the shorted input terminals which will flow through  the input protection resistors  (51K + 51K, R101 and R102 in the 3457A; 5K + 5K, R17 and R18 in the 3458A), input selection fets etc. And if the amplifier really does precede the AZ switch, as they postulate, then its DC offset would also remain uncorrected.

They also seem to have made a major error in assuming their tests exercise two voltage ranges (.3 and 30v) which allow the differences between 1x and 100x pre-amp gain to be observed. Hower the 3457A has a 1x range of 3V so both .3V and 30V ranges use 10x gain plus the 100x HV attenuator in the latter case. The same applies to their 3458A measurments where they chose the 1V and 100V ranges which again both use 10x amplification.

--- Quote ---For comparison, the case of range 30 V is similar but with a 100x larger scale, and is directly related to the case of range 0.3 V. The measurable quantity in both ranges is ?rst multiplied by one third through the same measurement channel in IVAAC. The multiplied quantity for range 30 V directly appears at the terminal P2P8 as vs 8 , while the one for range 0.3 V passes through the preampli?er of gain 100 before appearing there. By comparing both ranges, effects of the preampli?er on the readings are evaluated.
--- End quote ---

I'm not at all convinced at their conclusions but they must be very smart people so it's much more likely that this just demonstrates my ignorance.

Any thoughts?

David Hess:
I have not read the paper yet and it has been a while since I studied the 3458A design but a quick look indicates that it may suffer from charge injection when autozero is turned on.

Not long ago, I asked an HP/Agilent/Keysight sales engineer about this and she had no idea what I was referring to which to me indicates an institutional lack of knowledge on the subject.  She should have known about it whether current HP/Agilent/Keysight designs suffered from it or not.

Dr. Frank:
All chopper circuits suffer more or less from charge injection and ugly current spikes.
I just read two articles from our PTB, about a new, discrete chopper design, which eliminates these spikes, for an extremely noise free amplifier. They also compared several new commercial chopper amplifiers..
These spikes at first create bias currents, and second offset voltages, if they are non symmetrical.

The 3458A shows ugly chopper spikes, when you measure into high impedance sources.
I encountered that, when I compared the 1uA ranges, my  34465A in series with the 3458A, and the 1uA current source.. The 3458A always created an overflow on the 34465A, and with the digitizing option, I was able to make the 3458A chopper spikes visible, these had about 2.5uA amplitude, afair.

Same will happen on the DCV range, but I did not investigate that, yet.

By briefly scanning this article here, I understood its content in the same manner..



Comparing a voltmeter with an bare chopper amplifier can be tricky, as the voltmeters usually contain input protection that can contribute significant to noise. For the 3457 this a significant contribution to the input noise.

Some of the DMM input stages (looks like the 3457 does by adjustment of the precharge amplifier) use a kind of adjustment to reduce charge injection. This might be needed to get very low bias - though adjusting for lowest bias might be different from minimal spikes. The charge spikes could also depend on the DC voltage as charge injection usually depends on voltage.


[0] Message Index

[*] Previous page

There was an error while thanking
Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod