Null voltmeter

[notfaded1]

I'm kinda giving up on my Fluke 825AR... I replaced the reference tube because I couldn't do the setup calibration process and the user guide said to replace the reference tube (which I did and it didn't help)... so I decided to buy an 845AR instead.  The 845AR is in great shape but on lower ranges it doesn't stick on a reading and fluctuates back and forth the needle.  I found some of the original neon tube types used in the 845AR the NE-2U neon tubes.  I'm really into neon and finding the type isn't particularly easy because there are many types of NE-2 tubes and unless stated well.  I'm not sure if the different NE-2 are interchangeable in the Fluke 845 really?

I also in the meantime scored a dekavider RV622A with a rack built in 10V source including an additional divide by 100 passthrough.  I was amazed because last night I hooked it up to my 34401A on 6 digits with the slower 100PLC.  The divider was spot on all the lowest ranges including passed through the additional divide by 100.  Only the largest first decade was off by a mV on a few numbers.  I could dial up exactly on all but the first knob (still only off by 1 mV) any number on the 34401A.  Attached some of the original pics... I'll take some of my own later and post or maybe make a video.  I'm tempted to take apart the 10V source and see how it's built inside.  I wasn't really expecting the 10V source to be so good and also the dekavider.  When 3 or 4 of my voltage sources including my Fluke 343A all agree it gives me some confidence.  The 343A is pretty spot on too except the 300V and I think either the 600 or 900 range are off by just a small amount.  All the other decades on the 343A are spot on.  I wonder if there's some relationship between the 300 and the 600 or 900 range?  I mean where 100, 200, 400, 500, 700, and 800 are all fine.  I'll do the test again and post the results... maybe someone here has an idea why only a couple high ranges are off just a little.  It would be nice if it was just a cap that could be replaced.  I do have a DE-5000 and was planning on testing all the caps but since it's already almost exact on 95%+ of the ranges I'm a little hesitant to just start tearing the 343A apart.

Bill

[Echo88]

Regarding the F845AR please read the following: https://www.eevblog.com/forum/metrology/teardown-fluke-845aabar-tweaks-and-mods-(and-repairs)/msg2595273/#msg2595273
(unless you already did...)

[Echo88]

Due to the inherent difficulty of building a 8.5 digit DMM lets start with something more approachable 
Attached is:
-my take on a modern nullmeter as general schematic with lots of remarks
-a list of all (shown on respective manufacturer-website) single Autozero-OPs made by Analog, TI and some examples from Renesas/Microchip/Onsemi and two good JFET-OPs for comparison. The list isnt complete, may contain errors and unreasonable current noise datasheet values are colormarked.  I used the current noise measurement results from user chuckb for 3 OPs and commented them in the current noise column.

Inspirations did come from K155/F845AB/Datron 1045. I still need to read the AVM2000-document and see how they did it.
In my case i wanted to go with a pure analog and battery powered design, avoiding mains-powering and any µCs and DCDC-converters so far.
Want a graph or digital display? Plug the recorder output into your DMM6500/HPAK34465A or any other DMM/logging PC-interface.

All based on the idea that a AZ-OP or a JFET-OP should be suitable as a modern replacement for the nice discrete chopper-implementations of yesteryear that is...

Maybe someone already did current noise measurements on AZ-OPs apart from chuckb and wants to comment?
Debunking datasheets in this regard is a nice sport.

Hope for some nice discussions and hints for schematicerrors. 

[Kleinstein]

I have looked a little at the current noise of AZ OP -amps.
I got a crude measurement for the MCP6V51:  around 200-250 fA/sqrt(Hz). This sounds reasonable: a little worse than the comparable, but more expensive LTC2057.
I also had a look at the AD8628: the datasheet value is a bit optimistic, but the noise is still low (have to look up the old data).

For the bias current there some effect of the input impedance / input capacitance on the input currents. With balanced capacitance one has a good chance to get better than the listed typical calues.
The bias values from Microchip are using Bias = average input current and this way get to optimistic. One would need to use 1/2 offset + bias as a comparable value.

The JFET OP-amps are not really a reasonable alternative, if there is no extra automatic/reasonable fast Zero function.

The usual photovoltaic couplers provide very little power. This could be only enough for a super low power OP-amp.
Probably better to use a small 2nd battery (e.g. 2 or 3 x AA) if needed for the analog output. 
As a problem the analog coupler is unipolar only - so not really suitable for a signal around zero. So one would need something else there.

For a nullmeter with already a divider for most of the ranges I see little need for a higher supply and thus the bootstrapping. With battery supply it helps to keep the power consumption low.
It is more that one may want a choice of the amplifier:  low bias or low noise as separate versions.

[Echo88]

Yeah, the LMP2021-datasheet page 18 touches on the Cin vs. Ib-subject, i worry that different capacitive inputs would destroy the Ib-compensation-approach.
Just need to add switchable input compensation capacitors and autozero/autobias-compensation-switches...
I wanted to omit autozero/autobias-switches at least to get a continuous measurement device, where there are no selfcal-pauses during measurements if possible.

The IL300 can be configured for bipolar output with a current offset, yet will still have relatively high current consumption.
The suggested MAX409A has low enough quiescent current as a recorder buffer, but there might be better OP or ways.
Using 2x AA-batteries is pragmatic and long lasting, might be the easiest way indeed.

The bootstrapping is mostly used to keep the common mode stable and therefore hopefully the Ib (as done in the Datron 1045), but with varying Ib vs. Cin that is debatable.

I assumed temp-compensated (e.g. with a thermally attached compensation transistor) JFET-OPs should also be usable, but i guess the inherent 1/f-noise would necessitate a periodical autozero-cycle?

[Echo88]

Would about a chopped OPA140/matched JFET-difference amp?
http://www.janascard.cz/PDF/Design%20of%20ultra%20low%20noise%20amplifiers.pdf page 8
https://www.analog.com/media/en/technical-documentation/application-notes/an93f.pdf page 10 i would have expected less than his stated 500pA...

Also id reduce the voltage ranges to:
1kV, 100V, 10V, 1V, 100mV, 10mV, 1mV, 100µV, 30µV, 10µV, 3µV, 1µV...12 instead of 19 ranges, with smaller steps for the interesting voltages.
That increases the number of suitable rotary switches (whom also need to keep up with 1kV, meh)

Edit: Trying to make sense of the AVM2000-input section feels like im having a stroke, geez.

https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=&cad=rja&uact=8&ved=2ahUKEwjhysXu4KH8AhUBKewKHRFWA1AQFnoECAsQAQ&url=https%3A%2F%2Fwww.eevblog.com%2Fforum%2Fmetrology%2Fwhat-makes-the-avm-2000-nullmeter-have-such-low-input-noise%2F%3Faction%3Ddlattach%3Battach%3D1024710&usg=AOvVaw2omcFQs96N4PyXlkHzRPGO

[Kleinstein]

For the ranges I would consider spliting them to 2 switches.  One for the coarse 1:10:100:... part and than a seprate  1:3 or even 1:2:5 switch later on. With digital read out one would not need the 1:3 steps this is more something for an analog meter movement.

In the plan the capacitance at the input is quite large. This can cause a slow response if the source impedance is high. For a response that does not depend on the source impedance one would have less filtering at the input and than a later filtering stage (e.g. after the clamping to avoid long recovery times after an out of range signal.
Besides the normal RC response the capacitors can also add to a slow response from dielectric absorption. This is especially the case after large excursions. Settling to less than 1 ppm of prior voltage can take quite some time - more than the classical value of some 14 RC.



The balance of the input capacitance is mainly the question in the low capacitance range. So 10 pF and 20 pF can behave different, but 100 nF and 1 µF could behave similar (both are large).

A discrete build chopper could be an option, and in some circuits the OPA140 could be a suitable AC amplifier, other circuit may want more like a difference amplifier instead of an op-amp. The amplifier part is usually not the difficult part - the hard part is more the switching, especially when a low bias is aimed for. Here the integrated switches in the AZ OP-amp can usually get better matching.  One the other side the CMOS AZ amplifiers can't have large capacitors inside and the CMOS part tends to have more 1/f noise than JFETs. So the integrated chopper stabilized amplifiers often use a relatively high switching frequency, whilt a discrete build could use a lower chopping frequency. This can help a little with less current noise and give the option to use a higher capacitance but lower noise amplifier. Another advantage in a discrete build chopper is the possibility to separate heat sources from the critical section. I mainly see a discrete chopper something for the very low noise but higher bias range ( e.g. < 10 nV/sqrt(Hz)).

The input section shown is more for a low bias version, not for a very low noise version (more like nV meter). The resistors for the protection alone give quite a bit noise. 140 K of series resistance gives some 48 nv/sqrt(Hz).  A circuit for very low noise would need a different kind of protection - possibly to a lower maximum voltage.

[guenthert]

I like the idea of building a null meter (in fact, did so, even if not all that successfully in first attempt).  Also I think it is an attainable goal (and perhaps relevant as there don't seem to be any made commercially anymore and eBay (asking) prices went through the roof).

In any case, I think it's important to start with use case / requirements.  The AVM-2000's specs sure are impressive, but would be rather a stretch to replicate, I'd think.  Things which make it difficult are (of course) very high sensitivity (low noise), high impedance, isolated output (at least when staying all-analog), very high insulation to ground (battery only should get you easily to 100GOhm, beyond that it'll get tricky).  I'd like to see the use case and error-analysis first before fixating the requirements.  Is, e.g. a 1000V range truly useful today (we're not dealing with tubes anymore, are we?)?  If only ever 30V max are applied, then input protection could be simpler.

From my experience with building one and restoring (on-going) of a HP 419A, I came to appreciate the mechanical aspects.  The 419A is built like a tank: double walled (inner cage on LO potential, outer on GRD), low thermal inputs (regrettably not isolated outputs), quality, sturdy switches.  Their weak point, if one can call it that, is that their neon tube driven photocell chopper fails after a few decades (the neons wear out eventually, but the photo cells apparently become slow with age alone).  The amplifier actually still works fine.  The 419A is spec'ed for 0.3uV pk-pk noise (at 3s settling time to 95%), but I've seen (and read about) 0.1uV actual performance, which I think would still be a challenge when going with integrated OpAmps alone.  Also it doesn't drift (unlike my self-made attempt) or rather the remaining, minuscule drift is lost in the noise. 

[guenthert]

The bootstrapping is mostly used to keep the common mode stable and therefore hopefully the Ib (as done in the Datron 1045), but with varying Ib vs. Cin that is debatable.

I don't follow.  GND is on LO, so after the voltage divider common mode is minuscule and even poor OpAmp have CMRR of 100dB or more.  Common mode is least concern in a null meter, which only needs to indicate null accurately.  I wouldn't mind a gain error of a few percent.  Ib better be close to zero (back to requirements / error analysis), so close that we can't measure it precisely and is then allowed to vary a few % as well.

[Echo88]

Jep, the used prices for nullmeters are absurd now, which makes it more motivating to DIY one.
I consider the AVM2000 an interesting specimen, but nothing more so far.
The nullmeter i like the most is the K155: solid state design, long battery life, elegant and simple frontpanel design.
Specwise the DIY-variant should be at least as good as the K155 in every spec, especially noise and bias.

Of course the necessary voltage range (1kV) nowadays is debatable, but usage cases like F752A need 1kV and the DIY-variant should at least be protected against 1kV-accidents and ESD.
The mechanical guarded double wall construction, the high quality panelmeters and those sturdy rotary switches are really quite nice indeed...quite expensive when wanting to obtain them new though.

I attached the unraveled schematic of the analog input board of the AVM2000-board btw, as it was of interested to me how they did it.
Still dont know how often they switch the relays, they speak of matched pair relays for polarity switching...
Difficult to believe (at least for me) that they achieve the same low bias/noise-specs with high impedance source, while not polluting the measurement with DCDC-noise.

[guenthert]

The mechanical guarded double wall construction, the high quality panelmeters and those sturdy rotary switches are really quite nice indeed...quite expensive when wanting to obtain them new though.

I got a 2nd 419A (with dying neons) as parts donor, which I plan (someday) to use as basis for a self-made one.  Proper mechanics are far beyond my abilities.

I attached the unraveled schematic of the analog input board of the AVM2000-board btw, as it was of interested to me how they did it.
Still dont know how often they switch the relays, they speak of matched pair relays for polarity switching...
Difficult to believe (at least for me) that they achieve the same low bias/noise-specs with high impedance source, while not polluting the measurement with DCDC-noise.

On the interwebs someone complained about the acoustic noise of the relays, which made me think that they are used as low frequency chopper (but I might be completely wrong about that).

The AVM-2000 is big on filtering afaiu, with settling times up to 100s (!), so I'd think the noise from the digitial parts and DCDC converter is filtered out.

[guenthert]

Specwise the DIY-variant should be at least as good as the K155 in every spec, especially noise and bias.

I don't think the K155 (or any other null meter) specifies the bias current, even though it is of interest.  Anecdotally it is very low (less than 1pA).

[Kleinstein]

300 nV_pp noise is about 50 nV RMS noise. A response of 3 s to 95% should about correspond to bandwidth of some 1 Hz and thus an input noise of some 50 nV/sqrt(Hz).
Compared to the Fluke 845 the protection is quite a bit lower resitance (only some 16 K ohm of series resistance) and thus less noise possible.
There are quite some AZ OP amps with a noise of less than 40 nV/sqrt(Hz) and not that much input bias. A small input bias could be compensated. A possible candidate in this range is the AD8628  (some 22 nv/sqrt(Hz)) and a chance to get a input current in the 5-10 pA range, maybe with an optional added capacitor.
Getting a low bias may need selecting the AZ OP-amp or possibly a compensation through a high value resistor (usually more than the 47 M Ohm suggested in the plan).

From the schematics the Keithley 155 has a compensation for the bias current with 1 Gohm and some -150 to +500 nA adjustment range. So the actual bias current seen to be in the -200 pA range and can be adjusted to maybe the +-5 pA range depending on the pot quality and stability of the bias.

[Echo88]

Youre right guenthert, bootsstrapping seems unnecessary given the signal is +-1mV max and the other values such as CMRR/PSRR/gain accuracy arent that important for this application.
Bias current is indeed not specced in any nullmeter.
It appears to me that the whole AC/Normal Mode Rejection for mains frequency and above is mostly done in the chopper and not the input filter, is that correct?
The K155 states 100dB Normal Mode Rejection Ratio at 50Hz and above, while LTSpice gives a attenuation factor of about 400 at 50Hz for the input filter: 24k + 1µ + 22k + 1µ + 22k +1µ.
If the 50/60Hz-filter could be implemented after the main-OP one could decrease the input protection resistor further, allowing for lower noise results/more freedom in OP-choice, maybe with a nice high order sallen key filter as one in the AVM2000?
I attached the changed schematic with a changelog and the suggested Sallen Key filter.

Happy New Year btw. 

[Echo88]

Here are another two nullmeters, for those that are interested in the topic:
https://nvhrbiblio.nl/schema/Philips_PM2434.pdf PM2434 Already JFET-based, but only 10µV as smallest range.

Attached is the schematic of the Leeds & Northrup 9828 Null Detector.
Somehow i cant find the pdf-manual right now, so the schematic has to suffice for now. The most "modern" nullmeter i know, as it already uses op-amps. I own one, but the meter is broken and i need to repair it one day.
It only has 4 ranges: 2.5µV, 25µV, 250µV, 2.5mV...truly a nullmeter 

[guenthert]

[..]
Attached is the schematic of the Leeds & Northrup 9828 Null Detector.
Somehow i cant find the pdf-manual right now, so the schematic has to suffice for now. The most "modern" nullmeter i know, as it already uses op-amps. I own one, but the meter is broken and i need to repair it one day.
It only has 4 ranges: 2.5µV, 25µV, 250µV, 2.5mV...truly a nullmeter 

But the interesting stuff is still being done by discrete transistors and some transformer magic (wonder what the input impedance is).  I see it being referenced in '74, so not much younger than the Keithley 155.

A 2.5mV max range certainly makes input protection easier.


Happy new year everyone!

[Vgkid]

Attached is the schematic of the Leeds & Northrup 9828 Null Detector.
Somehow i cant find the pdf-manual right now, so the schematic has to suffice for now. The most "modern" nullmeter i know, as it already uses op-amps. I own one, but the meter is broken and i need to repair it one day.
It only has 4 ranges: 2.5µV, 25µV, 250µV, 2.5mV...truly a nullmeter 

Thanks. I own a modified version of a 9829 linear amplifier. It does work sometimes. It has been in storage for years , along with all of my gear.

[Kleinstein]

The L&N 9828 circuit looks rather low input impedance: the resistors in the protection are only 3x2.2 K and the capacitors are large.
It is strange that they use a BJT based amplifier after the transformer. Normally the transformer would increase the voltage and impedance and with a high impedance and AC only FETs are usually better.
There is some 6.2 M load to the AC amplifier that would transform back to the input.
At least they have a compensation for residual input bias, but the resistor value is not readable - likely in the Gohm range.

Another strange part is that the input capacitors seem to be electrolytic. For me this would be a big no go.

The PM2434 has the JFETs at the recorder output, not at the input. The input is rather similar to the K155, with MOSFETs for the chopper and a BJT based amplifier.  The output with rectifier and seprate polarity indicator is not that suitable for a null-meter.

Today a moder AZ OP-amp should be a good enough substiture for the input, if the requitements are not extreme ( bias < 2 pA or noise < 6 nV/sqrt(Hz)).
On chip they are likely quite good in compensating the switching spikes and charge injection.

[guenthert]

[..]
It appears to me that the whole AC/Normal Mode Rejection for mains frequency and above is mostly done in the chopper and not the input filter, is that correct?

I'll need to have another close look at the K 155's schematic.  The chopper eliminates offset and low frequency noise from the amplifier itself, but I'd think you want to filter "high" frequency (higher than chopper f. / 2) of the input signal before the chopper.  The demoduator is followed by a low pass filter which is meant to effectively filter the chopper f. (and higher) out.  To do that, the knee frequency will be considerably lower.  The line frequency is filtered then as side effect methinks.

As a side note:  HP's 419A uses a two stage RLC low pass filter before the chopper.  I couldn't find the specifications for the coils, but given that they're housed in a case like the one used for the neon transformer, I'd think they are rather large.

[guenthert]

The L&N 9828 circuit looks rather low input impedance: the resistors in the protection are only 3x2.2 K and the capacitors are large.
It is strange that they use a BJT based amplifier after the transformer. Normally the transformer would increase the voltage and impedance and with a high impedance and AC only FETs are usually better.
There is some 6.2 M load to the AC amplifier that would transform back to the input.
At least they have a compensation for residual input bias, but the resistor value is not readable - likely in the Gohm range.

I read R106 as 6*10^9Ohm 20%.

Another strange part is that the input capacitors seem to be electrolytic. For me this would be a big no go.

Perhaps they figure that at 2.5mV max, leak current through an electrolytic C. rated at 20V will be negligible (and only contributes to gain error anyhow).

[Echo88]

Regarding the 9828: The input caps are 8.2µF tantalum and the Ibias-compenation resistor is 5GR.
What are you thinking about the proposed active 4th order filter behind the main-OP to get the required NMRR at 50/60Hz while being able to use lower input resistance (at the expense of more quiescent current for the necessary OPs)?

The HP419A (~16kR input resistor) btw. only states max50Vin overvoltage on the most sensitive ranges, while K155 (24k input resistor) allows "momentary" overvoltage up to 1.2kV and the F845 (150k input resistor) 1.1kV apparently indefinitely.

I made some further changes to the schematic regarding the rotary operation switch and the battery charge level measurement.
As the nullmeter should be buildable by everyone i reduced the operating mode rotary switch poles x positions (like it did with the range switch, from 19 -> 12positions), to significantly reduce price and make it easier to obtain such switches from a wider range of manufacturers/market places.
It would have been nice to keep the elegant 2x rotary switch K155-frontpanel-design, but when i see on mouser that there are maybe some 20 fitting 8x5 rotary switches starting at 60$ then it would reduce the willingness of people to build the nullmeter.
Let me know what you think. 

[Kleinstein]

The idea with a 4 th order filter behind the main amplifier makes some sense. However one would still need enough headroom. A +-5 V for the main amplifier (e.g. a LTC2057) would help somewhat here. One may still want also some gain (e.g. a factor of 10, maybe even 100) after the filter. This would also help a little to not have that much gain (up to 1 million) in 1 stage.
The old null-meters also have some low pass filtering and DC gain after the chopper part. The should also be the main low pass filter. It does not look like a proper active higher order filter though. More like 1st order for the main part. The overall filter-function / response is hard to see (may want simulation) as it involves the chopper part and gain there.
´

A little more filtering at the input could still make sense to keep the amplitude of mains hum small, not to easily drive the main amplifier to limiting.
Unless AC is wanted I think there could be more filtering at the input, maybe some 100 nF and also filtering before the MOSFETs to add some ESD protection.
Very low resistance at the input would only be needed for a very low noise version. Unless really needed for the noise, more resistance in the protection is OK and helps.
The very low noise version would not be ideal for high impedance signals and thus does not need low capacitance.

The input protection in the lower version still looks a bit odd: claming to ground is still way better than clamping to the supply. The voltage is small anyway and limiting the voltage at the capacitors helps to speed up recovery from overload

The input filter in the K155 is more Sallen key like. This reduces the effective input capacitance. The first to capacitors are not towards ground, but towards the Feedback signal which approximates a buffered input.
For protection one may want extra clamping diodes in the path too.

Electrolytic capacitors have not just leakage, but also a leakage like current from dielectric absorbtion. The DA is usually assumed to be linear and thus also a problem with small voltages. In overlaod the voltage at the capacitors can be in the 1 V range and recovery after that to a low current could take several days.

[Echo88]

I integrated the suggested changes, especially splitting the gain between 2 OPs makes sense regarding GBW and mains hum saturation.
Why did you want to set the Sallen Key filter between the gain-OPs and not behind them?
I did some calculation based on the usage of the nullmeter for F752A and F720A as worst DUTs (respective output resistance).
Seems so far the thermal noise of the 70kR would dominate the calculation.
The AD8628 looks suitable for this task.
Are there any other suggested nullmeter use cases apart from resistance dividers and comparing voltage references, where an adapted nullmeter with lower noise could be more suitable?
I commented everything in the schematic.

[Kleinstein]

For the input protection I would consider to have the diodes for clamping a bit further down the line. Instead of 2 JFETs one could use a low leakage diode ( BAV199) they are cheap, small and may withstand a higher current.
The capacitors before the diodes should be small with a high voltage rating. Class Y rated capacitors (usually 4.7 nF) could be suitable. Otherwise PP type capacitors that are oftenavailable with good voltage ratings. The main capacitance should be at or after the diodes.  A MOV in parallel to the diodes makes no sense - it would never engage. If at all a MOV would be at the 2nd last stage to protect the capacitor.

The filter between the gain stages helps to make the circuit more resilient to hum. With the filter after the gain, hum could drive the amplifier into saturation as there is not much headroom.
This is especially a point with not that much filtering (smaller capacitors) at the input.

Not the main use, but as a good meter for low voltages one may also use the null-meter for other small voltages, like thermocouples. The AD8628 should still be good enough for this. For really small voltages there are special nV meters, but these usually come with a higher input bias.

The AD8628 has a max supply of some 5 V.
I have done a crude test on the noise current an got something on the order (not very accurate though) of 50 fA/sqrt(Hz).

The first amplifier stage should have some capacitance in parallel to the FB resistor. This can be part of the filtering and as a minimum be an option to effect the input bias.
Chances are one could get away without the bias current compensation with the AD8628 - still good to have the option.

[elecdonia]

Attached is the schematic of the Leeds & Northrup 9828 Null Detector. Somehow i cant find the pdf-manual right now, so the schematic has to suffice for now. The most "modern" nullmeter i know, as it already uses op-amps.

But the interesting stuff is still being done by discrete transistors and some transformer magic (wonder what the input impedance is).  I see it being referenced in '74, so not much younger than the Keithley 155.

Interesting: Traditional transformer-coupled chopper amplifier but with FET switches in front end vs. mechanical chopper. What do the two coupling transformers look like? And what are their specs: DC resistance & inductance of each winding, turns ratio?
Also what is chopping frequency? Mechanical choppers operated at 50, 60, 83, or 400Hz.