Can I auto-zero a sub-microvolt DC signal with a relay? (null detector)

[ckocagil]

I've studied a few DIY null detector designs so far, and I got to a point where I'm really wondering if a scheme like the following could work:

Take an SPDT electromagnetic relay and have it switch at something like 10 hz. Connect the relay such that its output goes into the non-inverting input of the opamp, while the input switches between the signal and ground. Have a decent ADC measure the amplified signal, get it into a microcontroller which would then report the difference between the high and low half cycles.

If this works, I would have the best of all worlds: low noise, sub-microvolt input offset and thermal EMF, low bias current (I'd use a JFET input opamp). It likely won't, which is why I'm posting it here so one of you can point out the glaring error.

[Kleinstein]

Mechanical relays tend to have problems with thermal EMF, as the coil needs quite some power for switching. At 10 Hz even a bistable relay needs quite some power.
A second problem is the contact forming at very low current - not all contacts work that well at very small currents. Gold or mercury whetted contacts are the normal choice for similar jobs - though hardly at 10 Hz.

The more suitable type of switching are electronic switches, like  JFETs, MOSFETs or PhotoMos couplers.

[thermistor-guy]

...
If this works, I would have the best of all worlds: low noise, sub-microvolt input offset and thermal EMF, low bias current (I'd use a JFET input opamp). It likely won't, which is why I'm posting it here so one of you can point out the glaring error.

Not sure about sub-microvolt, but the principle works. Leeds & Northrup used electromechanical choppers in its DC measurement and recording instruments 40 or so years ago. E.g. have a browse through:
https://f01.justanswer.com/QnsDptPr/PF.pdf

[TiN]

Here is how Keithley does it:



This is part of schematics from Keithley 7168 nanovolt scanner card, specified <30 nV offset per channel.

[guenthert]

Well, not just in pre-FET times, but also in the 21st century relays have been used for that purpose:
"Input signals are applied to the AVM-2000’s low-thermal EMF binding posts. From here the signals are
routed directly to the programmable input attenuator. Depending on the selected measurement range, the
input signal is applied directly to the input amplifier (100 nV – 1 mV ranges) via a matched set of lowthermal
EMF polarity reversal relays, or is attenuated to be compliant with the input amplifier’s dynamic
range and then applied to the input amplifier (3mV – 1000V). On all ranges, all stages of the input
amplifier allow sufficient head room so peak noise does not cause limiting and very high loop gains ensure
low-distortion. This allows the AVM-2000 to make full use of the digital filtering technology to eliminate
unwanted noise from the measured signals."
[https://www.tegam.com/wp-content/uploads/2015/11/AVM-2000-Manual-Rev-D-for-firmware-2.1.14-or-later.pdf]

[dietert1]

For the analysis of thin film resistor arrays i am using a relay scanner made of bistable relays with gold plated contacts. The coils of those relays are 180 Ohms and operate from 3 V. Current pulses are 5 msec, so there is no measurable thermal voltage as long as you operate a relay once per second or less (average coil power less than 0,5 mW).

We have lots of VXI switch boxes built with reed relays. Those are specified with "thermal voltage < 4 uV".

Regards, Dieter

[ckocagil]

What worries me about JFETs is that I don't know which parameters (if any) should match. Do I need a matched pair? Do I need low R_on? Are the JFETs going to introduce any voltage offset (except the thermocouples)?

Since I can't find much info online about solid-state switch selection for low DC voltages and I don't have any gear to run experiments with, I considered relays to be the more fool-proof way. Mercury wetted relays look harder to buy these days due to their mercury content. I'll look into gold plated relays.

This is part of schematics from Keithley 7168 nanovolt scanner card, specified <30 nV offset per channel.

Great article, thank you! Too bad it doesn't give me the exact JFET used in the Keithley. It likely would be out of production or my budget anyway. Do you have any resources on JFET selection for this purpose?

[Kleinstein]

A JFET should (no RF signal superimposed) not add any offset voltage except thermal EMF. The main parameters to look at are the R_on, leakage and maybe gate capacitance (as an approximation to get charge injection).

A classical choice for switching is 2N4392 or a modern SOT23 equivalent. These are for relatively low R_on and reasonably low leakage.
For a chopper there could be other better choices (especially lower voltage types) with less gate capacitance. A good resource on JFETs  the older Fairchild (now ON) application note AN-6609.
An important point is to keep thermal gradients small at the FETs.

As the voltage is expected to be relatively small, one could also consider a modern CMOS switch chip (e.g. DG4053).

Switching with 10 Hz sounds like relatively fast, especially for mechanical relays.

[thermistor-guy]

A JFET should (no RF signal superimposed) not add any offset voltage except thermal EMF. The main parameters to look at are the R_on, leakage and maybe gate capacitance (as an approximation to get charge injection).
...

Leakage will rise exponentially with ambient temperature, so be conservative if designing with room temperature leakage specs. Any ESD damage to the JFET will increase the leakage, so I suggest characterizing the leakage of the finished assembly at the upper ambient temp.

...
Switching with 10 Hz sounds like relatively fast, especially for mechanical relays.

Historical note: loop disconnect dialling for step-by-step relay-based telephone exchanges (1970s and earlier) used to use 10 pulses per second. The relay equipment of the time could detect and regenerate the pulses at that speed.
https://en.wikipedia.org/wiki/Pulse_dialing

[David Hess]

Chopper stabilization can be done with relays but there are lots of options including:

1. FETs, both MOSFETs and JFETs.
2. Bipolar transistors intended for chopping applications and with some cleverness, even normal bipolar transistors.
3. Photo resistors - these are one of the higher performance options.
4. Photo JFETs - I do not know how these work.

[Gyro]

Switching with 10 Hz sounds like relatively fast, especially for mechanical relays.

A lot of (well at least, some) Clare mercury wetted relays were rated to >100Hz.

Of course this depends a lot on whether this is a one-off experiment and probably the vagaries of ebay.

[voltsandjolts]

Why not use an auto-zero op-amp? Less than 1uV offset for less cost than any workable relay system.

[Vtile]

Rotating mercury switch. 

[Marco]

low bias current

For the lowest bias currents you want to precharge the input(s) of the amplifier, even with fA leakage CMOS inputs chopping will pull a DC current.

For IC designers mitigating that sort of thing just means throwing a few more square um at it, for you not so much.

[Vgkid]

Rotating mercury switch. 

I have seen that used in polarity reversal switches...

[ckocagil]

I played with some switch designs based on JFETs. A very good candidate seems to be MMBF4391/2/3 in plastic package. But the spec says a maximum of 100p leakage current and the LTSpice simulations using the official ONSemi model show a leakage current around 20-30p for my case. The current is pulled from the device being measured.

Is this too much current for a decent-ish nanovoltmeter that would affect my measurements?

My gut said yes, so I looked for some analog switch ICs. Analog Devices has these switch/muxes with only a few picoamps of typical leakage:



Their specs look incredible. It looks like I should be able to use these switches for a nanovoltmeter, perhaps even an electrometer/picoammeter. Am I overlooking something?

[niner_007]

COTO 3500 reed relays are rated at < 0.5uV EMF, but I believe you have to make a custom order, and they will be expensive, I expect around $60-$100 per relay

A mechanical relay doesn't make sense if you are going to turn it on and off all the time

[Kleinstein]

Testing leakage at the sub 10 pA level is slow and thus many parts have quite a spread from the typical to the maximum specs for leakage.
Using CMOS switches can be a good option. One point to watch for is the charge injection: this can also act similar to bias currents, especially if switching is relatively frequent. Charge injection also usually depends on the voltage level and the specs are sometimes confusing - especially 1st page number may be near best case. Low charge injection parts tend to use some compensation technique. So the current noise from variations in the charge injection can be higher than just looking at the charge injection and leakage numbers.

In a chopper application the voltage over the switches is usually small, so that the leakage may well be smaller than under the test conditions for the switches (quite often with a relatively high voltage).

[David Hess]

I wonder if anybody ever made a chopper or did low input current multiplexing with diode switches.

[guenthert]

I wonder if anybody ever made a chopper or did low input current multiplexing with diode switches.

  I didn't make one, but ran some simulations in LTspice in the context of a precision PWM voltage source.  It didn't seem to work well.  I attributed the failure in the end to the small, but not negligible (if switched with high frequency) capacity of the diodes.

[Kleinstein]

Diode switching is used in ring mixers. They are nice for RF, but I would not expect good performance at near DC due to thermal effects.  The diode drop is some 2 mV/K, while thermal EMF from dissimilar metals is more like 20 µV/K. The diodes also have quite some power dissipation compared to CMOS switches.

I used a ring mixer for essentially DC, and it worked better than expected.  Still  there was some low frequency noise in the µV range. I don't know the source,  but temperature fluctuations are well possible. A more classical chopper was not an easy option at 50 MHz. Due to thermal effects I would also expect some residual temperature drift, possibly not even better than a classical BJT based amplifier.

[iMo]

When talking mixers - long time back on the eve of modern SDR I used to use FST3253 switches as the mixer. A few ohm when on, pretty fast, 5V, no idea about other parameters, however.

[Conrad Hoffman]

You can get reed relays (search on reed contact) at about ten for a couple dollars. Add your own coils for a cheap experiment. They should be very fast.

[dietert1]

The proper technology is using opto fets as in later Fluke 845A null detector models, for example  ON Semiconductor H11F1M. Those parts are supposed to deliver clean signals up to some KHz without charge any injection.

Regards, Dieter

[David Hess]

The proper technology is using opto fets as in later Fluke 845A null detector models, for example  ON Semiconductor H11F1M. Those parts are supposed to deliver clean signals up to some KHz without charge any injection.

You could also *make* photoFET using a part like the Linear Systems LS627 or one of the parts from Crystalonics and the emitter of your choice.  I have seen some precision instruments that went this route with light pipes or just connecting the emitter and detector plastic tubing.