Null Detector based on Conrad Hoffman design check

[hozone]

Hello,

I would like to build a null detector based on the Conrad Hoffman design.
I've read the Conrad Hoffman article, find it here: http://conradhoffman.com/mini_metro_lab.html
And this interesting post here: https://www.eevblog.com/forum/projects/conrad-hoffman-null-detector/

The power stage it's built by using 4 AAA battery (6V), then a 5V LDO, then an op-amp as a rail splitter, so I'm going to use +-2.5V for the main circuit.
The op-amp I've selected is an AD8628, based on this @Kleinstein comment https://www.eevblog.com/forum/projects/conrad-hoffman-null-detector/msg3026766/#msg3026766
The only accurate resistor will be the 0.1% on the input divider.
Before going to test this circuit, do you think there is something wrong here?

Thanks

[Kleinstein]

The amplifier needs accurate resistors for both the input divider and the resistors setting the gain.
A Nullmeter is normally not about high accuracy in the scale factor, but good accuracy of the zero. So the resistor accuracy may not be that critical.

One may get away with fewer steps in the input divider (e.g. only 1:100 and maybe 1:1000). E.g. divide by 10 could be realized as divide by 100 and x 10.

I would make the input divider higher impedance, like 1 M if not 10 M.

The OP-amp for the rail splitter is not critical and should be a non AZ type.

[hozone]

Thank you!

I've just realized I've not attached the schematic. I've now edit my post, you can now find the schematic.

[iMo]

I would put a small resistor at the U2's and U3's output (like 33ohm or something like that).
Why such big 470nF in the feedback?
BTW., I had similar null meter in past with only one range (x1000 I think) and it covered all my use cases at that time..

[hozone]

Thanks @iMo
I get the reason after the U3, but even after the U2?

Another question is about the resistor accuracy, I've put 0.1% but maybe I can use less accuracy, cause on negative input of the op amp there are 1%, and they are trimmed by RV1, right?

I also would like to add an analog millivolt meter +-100mV on front on my case, switchable with banana plug.

[Kleinstein]

RV1 trims the offset, not the gain (only a very minor (e.g. 5 ppm) change in the gain).
With an analog meter movement an 1% or 0.5% accuracy could be good enough for the resistors. The analog movements are usually not that stable and a bit hard to read. An analog read out may want a trimmer for the gain. One could in this case consider an extra switch and series resistor for the movement to also get extra ranges as  x 3 or so - this could be even external to the PCB and still an option for later if wanted.
Today resistors with 0.1% are not longer that expensive and they may make some sense when using digital read-out.

A resistor (e.g. 50-100 ohm) at the output of U2 is needed, as most OP-amps do not like such a large capacitive load as the 2 x 100 nF shown.

The relatively large 470 nF capacitor at the feedback makes some sense to get some filtering action for the ranges with the highest gain. With a digital read out I would consider less capacitance there and also at the input.

The 1117 type LDO is one a kind of half way low drop type. There may be better ones to use more of the battery capacity (e.g. MCP1703 , LP2950,...).

[hozone]

Thanks!

I'm sorry but I don't get the modification I've to do for the analog meter, I've never use one analog. The series resistor and switch you are telling me are on the feedback network?

So:
- add 50ohm out of U2 and U3
- replace the LDO with something like TPMCP1703T-5002E

For digital meter usage:
- reduce C8, C9, C10 to 100n
- possibly use 0.1% for R5,R6,R7,R8 and R12,R11,R10. Just those ones?

For analog usage:
- all resistors can be 1%

[Kleinstein]

For the precision resistors one would have to include R9 too.

With the relatively low resistance of the divider the need for high accuracy is not that large. The 200 K resistor chain is quite some loading and this already limits accuracy a little.

The optional x3 range extension for an anlog meter would be between the shown circuit and the meter movement. A series resistor can reduce the sensitivity and this way extend the range by something like a factor of 3 to get a little around reading at only a small fraction of the range. Depending on the meter movement  (many come with a current rating, like 100 µA or 1 mA) one may need a resistor for the normal range too.  With a PCB this could be separate resistors for the output to the digital out and analog meter.

[Conrad Hoffman]

D3 provides protection one way but I think you need a second one for the other polarity?

[Kleinstein]

As shown in the plan D3 would only protect in one direct. However the normal version of the BAV199 is not common cathode, but 2 diodes in series.  The problem seems to be having the wrong symbol.

[hozone]

Thank you for clarifications!

I've modify the schematic:
- R5,R6,R7,R8,R9,R10,R11,R12 are 0.1%
- added 100ohm out of U2 and U3 (R16,R17)
- changed the regulator to the MCP1703 5V fixed version
- added a low battery indicator at ~5.3V almost
- updated BAV199 symbol and connections

I'll go for the digital one.

If this schematic pass your check I'll try the PCB prototype for this one

[Kleinstein]

Some reverse polarity protection at the supply may be a good idea, even of just a diode to short out a wrong polarity.
The low batt uses quite some current. With modern LEDs one could use less current and maybe use a slightly lower power version. The amplifier should not be that picky about the supply, so the limit could be quite close to 5 V.

The regulator may also want ceramic capacitor at the input side. It is at least good to have the option to add one.

[hozone]

Thanks!

- diode added

the low voltage supervisor is set to 5.3V cause the MCP1703 needs +0.3V in, so it's 5.3V

[retroware]

Have you given any thought to adding any sort of input bias current compensation? If one is using a null meter in combination with a high output impedance source such as the Fluke 720 KVD, then even the low bias current of the AD8628 can become significant - generating on the order of a microvolt or so of offset. Early meters such as the HP 419 didn't require this compensation given that they were using a pure resistive chopper.   Later meters like the K155 use a teraohm resistor to generate a small offset current. I've often worired/wondered about the drift in that part of the circuit.

[Kleinstein]

Adding a circuit for input bias compensation can absolute make sense. It is good to have at least the option - one can still decide not to populate a few reistors and the pot.
The input bias specs for the OP-amps are usually for the full common mode range, including the ranges close the the supplies, that can have higher bias. At a fixed center range and with well behaved source impedance the actual input current is often smaller than the spec limit, often even less than the typical value.
For use with a high impedance source it also makes sense to have a higher impedance for the divider - otherwise the divider would add to the noise.


For a digital read out it depends on the ADC used how much filtering capacitance makes sense.  SD or SAR (e.g. µC) versions can oversample the input relatively fast and near contineous. Here digital averaging / low pass filtering is preferred as it can be FIR type and thus fully settled in a limited time (e.g. 1 second or so). A classic dual slope ADC only average over part of the time (e.g. 1/4 for the classic ICL7106) and thus ideally would have quite some analog filtering in front for lowest noise.

[RolandK]

Power supply:
why not just use 4 eneloop divided in 2 groups. They deliver 2x 2.6V and and it's hard to get less noise. Just use a 2 pole power switch for positive and negative supply.

Circuit:
1.if you don't need high input resistance, you can use an inverting OP configuration. The operation point is then always at zero volts, so no common mode voltage and therefore better linearity.  (i think a µV meter is more useful then just a null-detector). Disadvantage of inverting configuration: low input resistance, e.g. only 10k if you have 10k input and 1 meg feedback resistor.
2. Use an OP with lowest input bias current, as it is drawn from the circuit to be measured. Input offset voltage may be trimmed.
3. Both inputs should see the same input resistance to prevent bias offset current. Beeing R1 input resistor and R2 feedback resistor at the - input, just parallel same resistors (1% is ok there) at the + input against GND.
4. Compensate bias offset voltage as proposed earlier. Just short the input and trim for zero output, leave it then switched on for 1 hour - your meter should not drift away.
5. The low pass filter at the input is against emitting chopper noise in the circuit to be measured, too. So adding a C against GND before the 2 20k may be a good idea.
6. Use a socket for your op-amp, so you can compare different types. Build it on a breadboard to see if it really works.

[hozone]

Thank you all,

so, I've a few questions. Please consider I'm more on the digital side, not so expert in analog.

@retroware, Kleinstein
Do you mean adding a voltage follower? Like some with an input resistor of 1k at +, and one on the feedback?

@RolandK
1. I'm going to use this for resistance measurement in Wheatstone bridge mode. Also I'm building this for experimentation.
2/3. You mean the MCP6001 used for virtual ground?
4. So, for calibration a) short +IN to GNDREF b) trim RV1 so that +OUT GNDREF is 0V c) check that there are not drift
5. Like another 100n, it should be 1500Hz cutoff when 1k (R8) is selected, 15Hz when 100k (R5) is selected

[Kleinstein]

The circuit as shown is a low voltage voltmeter / µV meter, not just a Null dedector. The main difference is that the null detectors don't care about the scale factor / accuracy and this way don't need calibration.

There is a balance between voltage noise on one side and input bias / current noise. So the choice of amplifier depends on the expected source resistance. The AD8628 is not the lowest bias, but usually still quite good and more like an intermediate choice.
For high source resistance, like 1 M and more I would prefer an max4238 or LTC2050.
This would only make sense when the input divider is also higher resistance (more like like 10 M or 100 M instead of the current 200 K).
Something like at least 1 M would already make sense with the AD8628, if not planing for high source resistance.
For a low source restance, if low noise is important an OPA387 / OPA388 can be better (but would need lower resistors in the protection to make full use of this).

The modern AZ OP-amps don't come in a DIP case - so a socket is not really a choice. It could also add thermal EMF and is not ideal.
To a large part this is a cheap circuit and would be more like building a 2nd version populated with a different OP-amp (and maybe different resistors) if needed.

The question is if one needs / wants compensation of the bias current. This circuit part may not be so obvious and there are different options (e.g. photocurrent, extra follower (e.g. MCP6001) from the inverting input of the AD8628 + shift + trimmer + a large (e.g. 1 Gohm) resistor to the input side). One would also need a higher resistance divider to "measure" the bias current.

Having the same resistance at both inputs of an OP-amp was valid for BJT based OP-amps to reduce the error from bias currents.
Things are diffferent for AZ amplifiers. Here the input currents are not equal on both side, but more of opposite sign. So adding resistance to the other side would only make things worse.
There is however a point in getting symmetry in the impedance at higher frequency (e.g. 1-100 MHz). In the curren plan the capacitors at both sides that make it low impedance on both inputs. Symmetry here can help with the switching spikes and charge injection related to this.

It makes no sense to switch to an inverting configuration. With low voltage DC this is a bad choice. If at all it may have some advantge with high voltage and AC and electronic range swiching, but these point's don't apply here.

[hozone]

Thank you @Kleinstein

I've read your post about the op amp suggestion on another topic in this forum, this is why I chose the AD8628

I'm going to use for precise resistance measurement so maybe I'm not going to need bias compensation. But maybe adding a simple bias compensation could help.

What do you think about implementing the simplest solution of a resistor followed by a trimmer from the negative input of the AD8628 to GNDREF? Should do the job?
If I've understand well, supposing a Ibias of 50fA (30 to 100 by datasheet AD8628). Vcomp = Ibias * Rcomp. If I want to compensate the voltage drop caused by the bias current of 1mV. Vcomp = Ibias * Rcomp. So Rcomp will be 20M.
Can i suppose 1mV or is too much?

[Kleinstein]

In may 2 tests so far the input bias with an AD8628 was quite a bit lower (more like 5 pA range including some CMOS switches in one case they could add or compensate). The specs may include the unfavorable regions close to the rails when the leakage current from the protection diodes no longer compensates. The input bias can also depend on the capacitance / impedance at the inputs.
With those low currents there is the problem that many parts are not tested very strict and the spec limits are more test limits not actual part properties. The typical bias is here the more relevant, but even that could be on the high side (e.g. including the worst case over voltage and possible improvements (tweak some capacitance / gate area) over time, not reflected in the DS).

A simple form for the input bias compensation would be a large resistor (e.g. 100 M, ideally more, but larger resistors get more expensive) from the OP-amps input towards a small voltage (some +-10 mV or so). Chances are one would have one of 2 resistor populated to have either a positive of negative compensation range and sign for the voltage.
This would add some loading (still small compared to the divider) and some biasing current (even if with a significant input voltage the overall current flows the other way). 10 pA would be 1 mV over 100 M ohm. 100 M would still be large compared to the resistors at the divider and not change the divider very much.

The bias compensation would only make sense if one can also test the bias and thus have a larger impedance (e.g. 1 M or 10 M) for the input divider. 1 pA with 1 M would still be only 1 µV and in the range of thermal EMF at the resistors. With only 200 K for the divder it does not need much thermal EMF at the divider to get more unwanted current.

Ideally one would do a polarity reversal for the excitation voltage in the bridge. This would average out both offset voltage and a bias current effect.

With high impedance sources in mind, one could consider using PP type capacitors for the filtering at the input. These are a bit larger, but have less dielectric absorbtion than the more common polyester film capacitors.

What type of ADC is planed for the display part ? This may effect how many ranges and what accuracy makes sense for the divider / amplifier.

[donlisms]

I think for balancing a resistance bridge, an analog null meter (with a needle) still makes a lot of sense.  I think it's harder to balance with a digital display. 

[Kleinstein]

For the display it depends on the needed bandwidth / speed that sets the noise level. With relatively fast cases the analog display can be faster. With a slow BW, which may be needed for lowest noise, a digital reading can be better. There is also the option to have digital reading via an ADC and still get a more analog like graphical display.

[RolandK]

@hozone: With 4 Eneloops you have +/-2.6V and GND ready, no virtual GND is needed.
 - on/off switch - +2.6V
|
Battery
|
Battery
|
 - GND=GNDREF
|
Battery
|
Battery
|
 - On/Off switch - -2.6V

As voltage surveillance you can add a low current LED on the supply side which draws less current. E.g. a red LED goes off at about 1.9 V. Trim the current consumption with a resistor that this side needs slighly more current.


@Kleinstein and Conrad Hoffman: Thank you for sharing your experience, allways a good read.

[hozone]

Thanks all,

I will use it mostly for resistance measurement (like the FIG7 sample here: https://conradhoffman.com/MML%20files/2_null_p5.jpg).I will not handle high impedance, and I stay with low bandswidth I suppose.
I'm going to use the accurate multimeter I've around here, that is a Brymen BM785.

@Kleinstein
Sorry for the dumbness of my replies.
Because I will not add large impedance on input divider, I does not need the bias compensation. To have the kind of compensation you are talking about (the one of the 100M to +-10mV) I have to build up a 10mV, right? That I don't have now.

@RolandK
That's a way. I can not find 1.3V cell, but rechargeable batteries are 1.2V, so maybe I can get it to +-2.4V. Find attached the schematic. Maybe that will be better than the 5V regulator + op amp virtual ground.
As for the monitor, trimming the resistor will not be easy, I'm thinking about a zener application, but the TL431 will cost almost the same.

[iMo]

With 2 packets by 6 position switch you may arrange it such you will measure the battery voltage in the first position, and then 100mV/10mV/..10uV ranges clockwise - as you usually start with the highest range (in reverse order as you have it today).