seeking advise and feedback on null detector

[NWerner]

Hello,

Some weeks ago I started work on a null-detector which was inspired by some old paper of Mr. Dauphinee.   You can find a conceptual sketch attached. Work was motivated to a large degree by my wish to get some more experience with magnetics.

It uses a capacitor to sample the input signal Uin. After some time the charge is dumped into a LC-tank where it hopefully starts a decaying oscillation. A transformer amplifies this oscillations voltage signal.

I added two complications:
a) oscillation is stopped at a multiple of the oscillation time period - so that in an ideal world a large fraction of capacitor charge is recycled -   
effectively bootstrapping the input.
b) to compensate for losses in the LC-tank I added some positive transformer coupled feedback

a first prototype was build. Besides fixing numerous silly bugs I learned:

- timing is more difficult than initially thought - so this probably needs a uC
- switches should use low rdson-types - 2N7002 are not appropriate
- mentioned complications seem to work.
- current signal (measured via inductive current probe) looks funny during turn-off. I guess its an artifact.

Does anyone know of similar circuits and can point to some literature or give some insights into limitations, and implementation traps?

Currently I tested the circuit only for large input signal  at high sample rates (so that operation of the circuit can still be monitored by my rusty analog scope). I intend to wander into mV-uV territory in the next version and changing sample rate to very low rates. Some feedback before doing so would probably be very helpful.

Thanks

[doktor pyta]

Very interesting.
Could You please share a reference of the the mentioned paper?

[RoadDog]

Take a look at Conradhoffman.com. He has a null detector built for use with digital meters.

Also take a look at the service manual for the Fluke 845ar. The smaller manual doesn’t include schematics or anything. The larger manual does. I believe you can download the big manual from Xdevs.com and a few other sites.

[RoadDog]

I took a look. Here’s where I found the full manual a while back.

http://nousnexus.com/Fluke%20845AR.pdf

[Atomillo]

Ive never seen a similar architecture before.
Could you please provide a reference to the original paper?
It seems quite interesting!

[NWerner]

Sure - Its:
"An Isolating Potential Comparator" T. M. Dauphinee 1953
Canadian Journal of Physics Vol. 31

figure 5 looks like steampunk-science.  But figure 9 is somehow familiar.
Please note, that this is only the original inspiration for my circuit - actual architecture differs quite a lot.

[Atomillo]

If I correct, can this be understood as a differential chopped amplifier with transformer input?

[Conrad Hoffman]

Yes, figure 5 is my favorite! This looks like something similar to what can be done with the LTC1043 or LTC6943 switched capacitor building blocks. There are time when complete isolation is needed, but also remember that this comes from a time when a good DC amplifier was just a pipe dream.

[Kleinstein]

It looks like a rather complicated way to build a chopper amplifier. I would not expect an expecially good performance. The input is somwhat double isolated, once with the transformer and than as an additionsl step from the flying capacitor. This could cause a slight problem, as the DC level for the intermediate part is not well defined.
The 2N7002 are not that bad, though low voltage fets can give a lower R_on with the same amount of gate charge.

If the turn off is not at a time with zero current, there will be an iduction spike and ringing at a higher frequency from parasitic capacitance / self resonance. So the exact timing could be important and effect the floating potendial from tiny changes in the charge injetion.

Chances are the damping reduction could be done from the main amplifier and not extra sense transformer(s) needed. This would be a little like old time regenerative / audion receivers.

[NWerner]

After several months I finally managed to proceed further. Notable advancements are:

 - another pcb revision was implemented to fix some bugs
 - a uC can now be used to control timing
 - a digital scope is used for observation of slow processes with low repetition rate
 - a fixed voltage source can be used to generate voltages from 100mV down to 100nV in some discrete steps and polarity.


First results were very disappointing however: Oscillation will only start for voltages <100mV!  Attached are screen-copies of oscillation
with emphasis on turn-on and turn-off behaviour. During turn-off  inductive peaking results as turn-off is not synchronized. During turn-on
the rather slowish  gate-drive seems to be problematic.

Two changes were implemented to achieve oscillation down to <100uV:
  - usage of large transformer
  - usage of mercury wetted reed instead of FET
I suspect, that both changes work by avoidance of ohmic losses during turn-on.

Current design is limited by poor rejection of 50Hz and incomplete software (not to mention ROHS, size, speed, huge cost and silly complexity for some very basic functionality)
Large transformer is not problematic for me. But merycury wetted relay is problematic due to ROHS, jitter and thermal emf.

For next revision I would like to reintroduce solid state switching (probably with FETs). I am however concerned regarding charge injection.
Can someone give some hints about how to avoid/compensate charge injection in this application?

Additionally I need some uC-based software to extract phase information from output signal in presence of frequency-shifts, noise and 50Hz interference. I am thinking of some kind of gated digital PLL.
Can someone point to some suitable algorithm?

[Kleinstein]

For looking at the oscillation, the phase of the oscillation and basic shape should be fixed (a 180 deg phase jump reflects the input sign). So one should be able to get the phase and frequency from a relatively large signal and than when looking for near zero just use correlation. So the signal is multiplied with an idealized reponse curve and summed up. With a zero mean for the ideallized curce there would be zero result if there is not oscillation and the response would be linear and near minimal noise. For mains hum suppression it would help to integrate over multiple such switching cycles in a way to have them equally spaced over the phase of the mains period and half of it. E.g. use 10 cycles that are 18 ms appart.

For a start I would consider one switching one side.