Null voltmeter

[zlymex]

The best range for Keith 155 is 1uV, 0.15uVp-p noise, 5 sec time constant, 1Meg input resistance.
The best range for Fluke 845A is 1uV, 0.2uVp-p noise, 5 sec time constant, 1Meg input resistance.
The best range for HP/AG 419A is 3uV, 0.3uVp-p noise, 3 sec time constant, 100k input resistance.
Normally, the usable resolution cannot exceed the noise, so 0.15uV to 0.3uV is the limit for these null meters.

If one DIY a simple null meter by chopper stablized opamps, the possible choice could be:
- LTC2054, very low typ Ib at 1pA, but not so low noise at 0.6uVpp(0-1Hz)
- LTC1052, very low typ Ib at 1pA, but not so low noise at 0.5uVpp(0-1Hz)
- LTC2057, low typ Ib at 30pA, low noise at 0.2uVpp(0-10Hz)
- ADA4528, not low typ Ib at 220pA, but very low noise at 0.1uVpp(0.01-10Hz)
Obviously I'll rule out the first two because of the noise. Also the large Ib and current noise of ADA4528(10pApp) will limit the input resistance to 20k or so that I cannot tolerate neither.
Therefore, I'll choose LTC2057 which has 1/4th of current noise density than ADA4528 allowing me to use 80k input resistor.

AVM-2000 on the other hand is far more better(by spec) than all the above: 100nV best range, 0.4nVpp noise, 1M to 1G selectable input resistance.

The use of Null meter for me could be:
- compare two voltages, such as two cells(1.018V nominal), two 10Vs.
- bridge balance detector.

Another special use of null detector for me is high-ohm bridge where a very sensitive current is required. I'll use ADA4530-1.

I've seen examples where ultra resolution is require such as transfer of JJVS where they use K2182A and EM N31(or alike) to measure the voltage difference.

[TiN]

There is also rainbow unicorn - preamp setup with Keithley 2001/2002 using variant of EM A10 (Keithley 1801).
DCV Ranges of such setup are 20uV,200uV and 2mV. I'll be getting back to this project soon.

[guenthert]

The best range for Keith 155 is 1uV, 0.15uVp-p noise, 5 sec time constant, 1Meg input resistance.
The best range for Fluke 845A is 1uV, 0.2uVp-p noise, 5 sec time constant, 1Meg input resistance.
The best range for HP/AG 419A is 3uV, 0.3uVp-p noise, 3 sec time constant, 100k input resistance.
Normally, the usable resolution cannot exceed the noise, so 0.15uV to 0.3uV is the limit for these null meters.

Is it?  Why is that?

If one DIY a simple null meter by chopper stablized opamps, the possible choice could be:
- LTC2054, very low typ Ib at 1pA, but not so low noise at 0.6uVpp(0-1Hz)
- LTC1052, very low typ Ib at 1pA, but not so low noise at 0.5uVpp(0-1Hz)
- LTC2057, low typ Ib at 30pA, low noise at 0.2uVpp(0-10Hz)
- ADA4528, not low typ Ib at 220pA, but very low noise at 0.1uVpp(0.01-10Hz)
Obviously I'll rule out the first two because of the noise.

Uh, I don't quite follow.  For many applications, e.g. buffering a reference voltage source to be used for a fast DAC, noise is an important limitation, but for a null meter?  As long as it is random (and only then call we it 'noise', right?) it averages out, doesn't it?

The use of Null meter for me could be:
- compare two voltages, such as two cells(1.018V nominal), two 10Vs.
- bridge balance detector.

Now, I've never had the opportunity to use one myself, but afaiu, those are (the) typical uses of a null voltmeter.  I'd think neither requires particular low noise in the amplifiers.  Well, if one depends on the analogue (moving coil) meter, than I can see that excessive noise makes measurements difficult, but these days we connect an ADC and calculate the average, no? 

[zlymex]

Uh, I don't quite follow.  For many applications, e.g. buffering a reference voltage source to be used for a fast DAC, noise is an important limitation, but for a null meter?  As long as it is random (and only then call we it 'noise', right?) it averages out, doesn't it?

Because it is a very low frequency noise of DC-10Hz, averaging takes long time say 10 seconds, and the 0.1Hz noise will interfere with it making the averaging even longer, and even lower frequency noise(0.01Hz, 1mHz, or lower) becomes troublesome.

The good news is that there is virtually no 1/f noise for chopper stablized opamps, the noise density remains constant when frequency goes down to 0.1Hz, or even below 0.01Hz, making it a good choice to DIY a simple NULL meter.

Two charts attached that can be regarded as the readout from oscilloscopes, one with 1/f noise and the other without.

[d-smes]

Two charts attached that can be regarded as the readout from oscilloscopes, one with 1/f noise and the other without.

The upper chart of AD8676 looks to have about 100 nV pk-pk noise while the lower LT2057 has 200 nV pk-pk noise.  Both have random "spikes" in voltage while only the upper plot appears to have shifts in the running average noise voltage (hard to tell with different vertical scale factors).  Is that the 1/f noise characteristic to which you refer?

Perhaps a related question- How can LT claim the noise voltage plot is representative of a DC - 10 Hz bandwidth with just a 10 second recording?

[zlymex]

Two charts attached that can be regarded as the readout from oscilloscopes, one with 1/f noise and the other without.

The upper chart of AD8676 looks to have about 100 nV pk-pk noise while the lower LT2057 has 200 nV pk-pk noise.  Both have random "spikes" in voltage while only the upper plot appears to have shifts in the running average noise voltage (hard to tell with different vertical scale factors).  Is that the 1/f noise characteristic to which you refer?

Perhaps a related question- How can LT claim the noise voltage plot is representative of a DC - 10 Hz bandwidth with just a 10 second recording?

Broadly speaking, yes. It's getting late here so I'll say some conclusions first and will explain why tomorrow.

The chart above for AD8676 is the noise for 0.1Hz to 10Hz bandwidth. If we modify the the bandwidth as 1mHz to 0.1Hz, or modify the the bandwidth as 0.01mHz to 1mHz, the chart will be the same(similar shape, same 100nVp-p). This can go on forever and theoretically these noise(of limitless bandwidth) will be added up and there is no boundary, and you cannot claim anything for bandwidth of DC-10Hz, that is the characteristic of 1/f noise.

However, for LTC2057, Vp-p will be dropped(to 1/100th or 1/10th?) if bandwidth is 0.001Hz to 0.1Hz, and continue to drop for lower frequencies bandwidth.

For the second question, for an opamp without 1/f noise, the Vp-p contribution from low frequency has a limit/boundary, the Vp-p noise for 0.1Hz-10Hz is not much smaller than 0.001Hz-10Hz which virtually the same as DC-10Hz.


Edit: From wiki: https://en.wikipedia.org/wiki/Pink_noise
'There is equal energy in all octaves (or similar log bundles) of frequency.'

In another word, all frequency bandwidths of below carry equal energy:
1-10Hz, 0.1-1Hz, 0.01-0.1Hz, 0.1m-0.01Hz, 0.01mHz-0.1mHz,,,,,, 0.01uHz-0.1uHz,,,,,
Therefore, in theory, there is infinite amount of energy in DC-10Hz spectrum for 1/f noise source.

[zlymex]

I found out a chart that I recorded before which can get an intuitive view of the LF noise of one of my Fluke 845AB.

[uncle_bob]

.......

In another word, all frequency bandwidths of below carry equal energy:
1-10Hz, 0.1-1Hz, 0.01-0.1Hz, 0.1m-0.01Hz, 0.01mHz-0.1mHz,,,,,, 0.01uHz-0.1uHz,,,,,
Therefore, in theory, there is infinite amount of energy in DC-10Hz spectrum for 1/f noise source.

..... but only if you have an infinite amount of time to watch the power meter ....

One would guess that someday there will be a paper that changes the way we look at 1/F noise and "infinite energy". In a bit over a century of study, it holds up so far.

Bob

[zlymex]

In another word, all frequency bandwidths of below carry equal energy:
1-10Hz, 0.1-1Hz, 0.01-0.1Hz, 0.1m-0.01Hz, 0.01mHz-0.1mHz,,,,,, 0.01uHz-0.1uHz,,,,,
Therefore, in theory, there is infinite amount of energy in DC-10Hz spectrum for 1/f noise source.

..... but only if you have an infinite amount of time to watch the power meter ....

True. That is just another way of saying: averaging is no good for lowering 1/f noise no matter how long you are trying.

[uncle_bob]

In another word, all frequency bandwidths of below carry equal energy:
1-10Hz, 0.1-1Hz, 0.01-0.1Hz, 0.1m-0.01Hz, 0.01mHz-0.1mHz,,,,,, 0.01uHz-0.1uHz,,,,,
Therefore, in theory, there is infinite amount of energy in DC-10Hz spectrum for 1/f noise source.

..... but only if you have an infinite amount of time to watch the power meter ....

True. That is just another way of saying: averaging is no good for lowering 1/f noise no matter how long you are trying.

Hi

Which brings us right back to the start of this ... why 1/f noise is an issue you *do* need to worry about. Chopper front end ADC's are indeed a good idea to the extent they can suppress some of the 1/f noise issues.

Bob

[zlymex]

Some null meters(of existing and DIY) do show worrying 1/f noise.

[Conrad Hoffman]

I've got a couple HP 419 meters and a Fluke 845. The 419s are good, but mine both need service. The 845 is my favorite, but watch out for leakage to ground. I think there's a leakage test in the manual involving high voltage. When I did it, a hard to reach piece of coax behind the front panel flashed over. I fixed it, but the leakage is still just out of spec. The mechanical design of that thing is, IMO, just plain stupid when it comes to service. I replaced my batteries with a big cap and zener diode, as they seemed to go bad on a regular basis. The null meter from the original MML articles worked well, but I just vastly prefer a center zero mechanical meter in this application. A couple of my bridges use eye tubes and as antiquated as they are, they make great null meters. Analog motion, but zero friction!

[guenthert]

A couple of my bridges use eye tubes and as antiquated as they are, they make great null meters. Analog motion, but zero friction!

I've seen magic eye tubes on radios (utterly mesmerizing), but never on test gear.  You wouldn't happen to have a picture?  Null meters benefit from battery operation, I'd think.  How did they feed the heater current for those tubes?  And don't use magic eye tubes a gate voltage in the order of ten volts?  How was the signal so far amplified?

[Squantor]

I've seen magic eye tubes on radios (utterly mesmerizing), but never on test gear.  You wouldn't happen to have a picture?  Null meters benefit from battery operation, I'd think.  How did they feed the heater current for those tubes?  And don't use magic eye tubes a gate voltage in the order of ten volts?  How was the signal so far amplified?

I have a Wayne Kerr bridge that uses a magic eye as for nulling the bridge and the Philips GM6020 used two indicator lamps to show the null. Philips used a second moving coil meter as  polarity indicator on many of their multimeters that you also can use as the Null indicator.

A couple of my bridges use eye tubes and as antiquated as they are, they make great null meters. Analog motion, but zero friction!

Ah that is a good reason to use those, I always wondered why they used a magic eye on the older bridges or millivoltmeters.

[Conrad Hoffman]

Here you go- pictures of a GR bridge using the eye tube:
http://conradhoffman.com/GR1611A.htm

[Kleinstein]

There are mechanical meters without a classical bearing and thus (static) friction to. So friction itself is not the problem with these, but response is slow and tends to show overshoot. You usually need to add dynamic friction to make is useful.

[Echo88]

Sorry for unearthing this thread, but maybe we can discuss the following question here:

I often read the claim that a Nullmeter like K155 or Fluke 845AB have the property of an infinite input resistance when Udiff=0 at its input-posts.
Since the datasheet of both devices mention input resistances of 1Megohm for the K155 and 10Megohm for the F845AB and one can see those resistors across input-posts in the schematic that claim is false.
I understand that if Udiff=0 across say a wheatstone-bridge then theres no current flowing at all through the nullmeter and therefore it appears to have infinite impedance or as if it wasnt in the circuit at all, but thats no intrinsic property of the nullmeter at all.
Therefore it should be possible to use any suitably high resolution, low noise and low bias current DMM as a nullmeter, as long as the bias-current from the meter doesnt introduce new errors due to high bridge-resistance (DMM-Mains-connected-common-mode-errors left aside here).
Fluke mentions this in an appnote, but also states that Nullmeter-input-resistance is infinite at null, which i dont understand. -> http://assets.fluke.com/download/calibration/seminars/replacinganalognulldetectingmeters.pdf Page 30

Can anyone point to my error in thought or is the appnote wrong in claiming "Input resistance infinite at null"?

https://doc.xdevs.com/doc/Keithley/155/29031D%28Model55%29.pdf K155 Manual
http://ftb.ko4bb.com/manuals/89.12.168.225/Keithley_155_Schematics.pdf
https://xdevs.com/doc/Fluke/845AB/fluke_845a_ab_sm.pdf Page 48

[TiN]

Perhaps that's due to no current flowing into floated nullmeter when there is no potential difference between the inputs = zero current gives infinite calculated resistance. That does not account for leakages, obviously.

[Echo88]

Interestingly the HP419A actually claims the infinite input impedance is achieved at null, because it uses a special battery-driven supply to completely compensate input current.
Since K155 und Fluke 845AB dont claim (or i didnt see it in the manuals) to achieve infinite impedance at null and its visible that theres always an input divider at the binding posts.

Ill assume the HP 419A (and maybe other types) is the basis for the claim that Nullmeters always have intrinsic infinite input impedance when nulled.

http://bama.edebris.com/download/hp/419a/HP%20419A%20DC%20Null%20Voltmeter.pdf HP419A manual, Page 6 specs

[Kleinstein]

The HP419 uses a LDR type optical chopper. So it is not even really high impedance, though possible a relatively low bias.

Part of the confusion is due to different definitions for the input resistance. One can use a differential resistance (dU/dI) and an absolute resistance U/I that can depend on where to have the zero point of the voltage.

The infinite resistance might be seen to describe zero bias or no disturbance. This is more something from the times of passive analog meters.

The differential resistance of the null-meter is still the same, no mater what voltage is measured.

[guenthert]

Sorry for unearthing this thread

No need.  It's my favorite topic, I got a little obsessed with these devices.

I often read the claim that a Nullmeter like K155 or Fluke 845AB have the property of an infinite input resistance when Udiff=0 at its input-posts.

That's bit sloppy terminology or marketing-speak.  What is meant is that, when null'ed, there is no loading of the source, i.e. no input current into the null-voltmeter.  Of course, the input current isn't exactly 0, but very, very small (low pA or high fA range, compare with the input current in the high pA range of modern long scale DMMs -- and that's average; their "pump out" current is pulsed and peaks in the nA range).  For old galvanometers it's obvious why there was no current at the null point (if I'm not mistaken, mirror based galvanometers had a sensitivity in the low nA range).  All modern null meter use however an amplifier and hence there's always some small input current (and for choppers this nasty pump-out current).  One can either put a lot of effort into minimizing such or - as HP and others did - add a burden voltage generator [bucking voltage is the proper term].  That burden [bucking] voltage is manually adjusted with the input shorted so, that the current across the input resistor matches and opposes the input current of the amplifier -> presto, no load of the DUT.

Away from null (|Udiff| > 0) the input current would be |Udiff/Ri| > 0, but since very small voltages are typically measured with a null voltmeter, only small currents flow.  Of course, the smaller the input resistance, the greater the error.  Unfortunately one cannot make the input resistance arbitrary large, as it is a noise source, which ultimately limits the sensitivity of the meter.

No (or very little) loading used to be important when using standard cells, as they had a fairly high output resistance and held only a small charge.  It's still of interest when using a bridge to compare very high value resistors or when using a voltage divider (including the still current Fluke 752A) to determine a voltage with high precision ("potentiometer method").

[guenthert]

Therefore it should be possible to use any suitably high resolution, low noise and low bias current DMM as a nullmeter, as long as the bias-current from the meter doesnt introduce new errors due to high bridge-resistance

One essential feature of null-meters is the very high (iirc Fluke claimed 1TOhm for the 845A) resistance to ground.  That's at least an order of magnitude higher than any long scale DMMs I'm aware of (there are digital electrometers with high impedance to ground, but they are typically not very sensitive) and could be of importance when measuring (or comparing) sources of high potential above ground with large output resistance, e.g. a bridge for large value resistors.

[mikron]

Perhaps that's due to no current flowing into floated nullmeter when there is no potential difference between the inputs = zero current gives infinite calculated resistance.

Following this kind of logic even the resistance of a 10 Ohm resistor rises to infinity whenever it is out of circuit.
    

[jfphp]

The AVM 2000 is the only modern null voltmeter but it is very sensitive to static discharge and cannot withstand 50V like the 419A on the most sensitive ranges. An interesting approach would be to copy the very simple schematic diagram of the null meter incorporated into the Valhalla 2720GS (a brilliant but largely unknown 8 1/2 digit DC calibrator).
I guess, I have seen on the net an upgrade of the HP 419A with a solid state chopper : any idea ?

[Kleinstein]

Away from null (|Udiff| > 0) the input current would be |Udiff/Ri| > 0, but since very small voltages are typically measured with a null voltmeter, only small currents flow.  Of course, the smaller the input resistance, the greater the error.  Unfortunately one cannot make the input resistance arbitrary large, as it is a noise source, which ultimately limits the sensitivity of the meter.

The input resistance of an amplifier is not directly producing noise. Good amplifiers have a noise level that is lower than the noise of a resistor as high as the input impedance. However there is still a relation that high input impedance amplifiers have high noise. Otherwise one could just add more in parallel and reduce the noise without a negative effect (except cost and power).
A different point is the resistance used in the protection at the input - this adds to the noise. At the very lowest noise level this can be a reason not to have so much protection.

No all modern null meters use an amplifier at the input. The most sensitive ones are superconducting ones that measure the current (down to the sub pA level) through a superconducting coil from it's magnetic field. These are thus null meters with a crazy low input impedance - just a few µHs and 0 Ohms, and they are crazy sensitive, like nV full scale and down well below pV.

So there is likely not just one ideal nullmeter that fits all tasks - it is still a balance between low bias or low noise.