Forums

Low noise, high bias voltage on picoAmp TIA's input, howto?

Started by timo...@ibtk.de May 21, 2021
Dear all,

at first I want to thank all the experienced guys, Phil, Winfield, Gerhard... who share their valuable knowledge. That helps a lot! Thank you!
I'm working at a Helmholtz facility in Germany/Dresden on equipment used in fundamental research.

To measure e- and ion beam currents down to sub-pA I built up Fameio, a module including TIA (9 decadic ranges), diffamp, ADC/DAC, supplies, bias -200V to +200V:
https://ibtk.de/project/hzdr/Fameio-presentation/20210521_Fameio.html
The variable 2mA current limited bias, which has to sit on the measuring tip (in relation to GND), is a deadly for the TIA input in the case of Triax cabling errors (shorts).
The TIA is very sensitive to bias noise, if the bias node is not perfectly AC-coupled to the front-ends OPA supply. As well as any differences between its +in and Guard_ext couple into the measured signal due to the cable capacitance.
For lowering the noise I will throw out the iseg modules and use a Royer.

Question:
I'd like to have a more robust bias-solution.
I thought about photocouplers to feed the DC voltage in series to the TIA tip directly, it's open voltage is nice (30V) but the current is way too low (Isc=4uA).
It could be possible to use an DCDC instead the same way, but it may be challenging due to the needs of isolation, screening and guarding.

Any idea would be appreciated.
Thank you very much!
Cheers, Timo
On Friday, May 21, 2021 at 10:08:25 PM UTC+10, timo.k...@ibtk.de wrote:
> Dear all,
> For lowering the noise I will throw out the iseg modules and use a Royer.
If you want a DC-DC converter, think about using a a Baxandall reseonant Class-D inverter rather than Royer. The problem with square wave inverters are the switching spikes, and Baxanadall resonant inverters switch more smoothly. You still have to worry pumping capacitative currents across isolation barriers, but there's less high-frequency trash around. Peter Baxandall invented the circuit back in 1959, but he published it in a British journal and Americans didn't know about it. Jim Williams eventually picked it up and ran with it, but clearly didn't know where it came from. He first mentioned it in the Linear Technology application note AN-45 which is dated as June 1991, and but AN49, AN51, AN55, AN61, AN65 concentrate on using it to generate high voltages to drive cold-cathode back-lights. http://sophia-elektronica.com/Baxandall_parallel-resonant_Class-D_oscillator1.htm has a link to Peter Baxandall's original paper, and some of my own variations on the circuit. -- Bill Sloman, Sydney
Hello,
Bill Sloman schrieb am Freitag, 21. Mai 2021 um 15:25:53 UTC+2:
> If you want a DC-DC converter, think about using a a Baxandall reseonant Class-D inverter rather than Royer.
Oh yes, I did not know about Baxandall, nevertheless I used it already (and mixed the names, excuse me, please). A magic circuit, started below 1V(!) and worked nicely to 30V on my bench... Please see the updated link, I added it's first simulations and measures (in German): https://ibtk.de/project/hzdr/Fameio-presentation/20210521_Fameio.html The AN118 helped a lot. Thank you for the additional hints! I did not know the great AN65 for example. One additional restriction is the allowable height below 10mm on the module's bottom side, so the core has to remain small.
> You still have to worry pumping capacitative currents across isolation barriers, but there's less high-frequency trash around.
If the circuit stays inside the module, these currents are not harmful. The bias generation is GND-related. It will be a problem, if located outside, in series to the TIA input. So the first option still seems to be easier. I'll have to go through the information you provided, thank you! Cheers, Timo
On 21/05/2021 22:08, timo.k...@ibtk.de wrote:
> Dear all, > > at first I want to thank all the experienced guys, Phil, Winfield, Gerhard... who share their valuable knowledge. That helps a lot! Thank you! > I'm working at a Helmholtz facility in Germany/Dresden on equipment used in fundamental research. > > To measure e- and ion beam currents down to sub-pA I built up Fameio, a module including TIA (9 decadic ranges), diffamp, ADC/DAC, supplies, bias -200V to +200V: > https://ibtk.de/project/hzdr/Fameio-presentation/20210521_Fameio.html > The variable 2mA current limited bias, which has to sit on the measuring tip (in relation to GND), is a deadly for the TIA input in the case of Triax cabling errors (shorts). > The TIA is very sensitive to bias noise, if the bias node is not perfectly AC-coupled to the front-ends OPA supply. As well as any differences between its +in and Guard_ext couple into the measured signal due to the cable capacitance. > For lowering the noise I will throw out the iseg modules and use a Royer. > > Question: > I'd like to have a more robust bias-solution. > I thought about photocouplers to feed the DC voltage in series to the TIA tip directly, it's open voltage is nice (30V) but the current is way too low (Isc=4uA). > It could be possible to use an DCDC instead the same way, but it may be challenging due to the needs of isolation, screening and guarding. >
I don't think connecting the bias voltage in series with the input pin is very practical - the voltage source would have to be very small physically in order to avoid excessive capacitance. If a transformer-based DC-DC converter were used, it would also be hard to get the inter-winding capacitance small enough to not suffer from interference from the AC voltage on the windings, even with quite careful screening. A photovoltaic device (perhaps driving a step-up autotransformer) might be ok, if it is small enough, but I think your existing approach is better. Anyway, I don't think that putting the voltage source in series with the input terminal has any advantage for the robustness against damage when the input is short-circuited, as the bias will still appear at the amplifier input. To make it more robust, I think you need to add more series resistance (perhaps with some measures to stop that from causing degraded performance) and/or add more protection diodes (which should be chosen carefully and operated at zero bias). From what I can see, it looks like you have floated the entire front-end at the bias voltage, including the shield cans (which would minimise collection of current from ions generated in the air inside the shield can). This is what I would have done. I presume that the power supplies for the op-amps are provided by DC-DC converters with their outputs referenced to the bias voltage. I could not find any diagram of the front-end that included the isolated power supplies, and the power pins of the op-amps seem to be hidden so I can't really tell what you have connected them to. It seems to me that you would need an instrumentation amplifier to shift the signal to be ground-referenced. I guess U7A is doing that, but I don't know what part number U7A is as it isn't labelled. If it is a difference amplifier then I can't understand why you connect the feedback resistor to its output (ground-referenced) rather than to the output of U5A. I do wonder whether it is worth considering replacing the switchable feedback resistors by a p-n junction (a diode-connected transistor, JL and PH could suggest one with low leakage and good log-conformance) and then digitising the result of that (as well as a replica diode at the same temperature and known current). That way you get a voltage proportional to kt/q log the current ratio. Whilst I would not expect much better than 5% accuracy that way, it might still be better than the tolerance of a 1 Teraohm resistor, and removes the need for auto-ranging, relays etc.
Hello Chris,
thank you for the hints!

Chris Jones schrieb am Montag, 24. Mai 2021 um 11:59:35 UTC+2:

> If a transformer-based DC-DC converter were used, > it would also be hard to get the inter-winding > capacitance small enough to not suffer from interference > from the AC voltage on the windings, even with quite careful screening.
Here are some interesting ideas to minimise the coupling capacitance: https://www.eevblog.com/forum/metrology/power-supply-for-voltage-references/
> ...floated the entire front-end at the bias voltage, > including the shield cans...
Yes, at first. But it measured ungly, maybe due to the "non-optimum" layout. So I put the inner shield back to GND with reasonable results - on the bench. Yes, the bias drives the common and the supplies of the floating DCDC used for all the OPAs related to the TIA, to bee seen at p. 13 of the document: 20210316_SuperSIMS_FAMEIO_MV40-Ersatz_Praesentation_Kirschke.pdf. The U7-A at p10 is an INA149, a TI HV-Diffamp. I like to simplify the structure a bit, omitting the composite amp. It's power is GND-referenced, of course. But it's input draws already ~0,5mA from bias and ~0.5mA from the TIA's output. Because the bias current possibly available at any in/out of the module must be limited to remain <3mA under all conditions for safety purposes, all power sinks should be kept at minimum. Yes, the INAout-connection is a mistake, it is really connected to R107 (U6-B: composite, to bring the voltage to +/-10V inside the loop, lowering the noise a bit). As well as the schematic was not actualised after correcting all the failures. Excuse me please. The schematics, layouts and production are done by our technicians, sometimes the information flow is not good enough. So my second job for troubleshootimg remains necessary. ;)
> ...replacing the switchable feedback resistors by a p-n junction...
Our "customer" physicists asked for an idea of such a version, so I simulated one, see 20210522_FAMEIO_SuperSIMS-MV40-Ersatz_S45-46.pdf which document was added to the link above. But it's still not realised by them, will take some additional weeks. For our purposes we'd like to get some more reliable accuracy at least at the higher ranges. We bought the resisors at SRT-Restech, the 1T in 2512 has 5% TK250, you're right. But the range _could_ be calibrated digitally, if necessary. But the log-idea itself is really good, i believe. On the other hand I do not have any practical experience with logamps... Thank you! Cheers, Timo
On 24/05/2021 21:14, timo.k...@ibtk.de wrote:
> Hello Chris, > thank you for the hints!
Sorry I didn't have time to read your documents more thoroughly before replying so I missed some things you had already provided.
> > Chris Jones schrieb am Montag, 24. Mai 2021 um 11:59:35 UTC+2: > >> If a transformer-based DC-DC converter were used, >> it would also be hard to get the inter-winding >> capacitance small enough to not suffer from interference >> from the AC voltage on the windings, even with quite careful screening. > > Here are some interesting ideas to minimise the coupling capacitance: > https://www.eevblog.com/forum/metrology/power-supply-for-voltage-references/
Thanks! I have to do something like this, I would use two cascaded toroidal transformers, on opposite sides of a metal wall with two holes in it, each toroidal core having its axis of symmetry parallel to the plane of the wall. The first transformer would have many turns primary : 2 turns secondary, and the ends of the 2-turn winding would pass through the holes in the metal wall. On the other side, there would be the other transformer with 2 turns primary to many turns secondary. The centre-tap of one 2-turn winding would be grounded to the metal wall so that the wires passing through the wall have equal and opposite voltages on them. Even with this arrangement and with PTFE insulated wire, I would not expect it to be useful below nanoampere signal levels unless I could arrange for the transformer operating frequency to be so far above the frequency of any wanted signals that it could be filtered out very well.
> >> ...floated the entire front-end at the bias voltage, >> including the shield cans... > > Yes, at first. But it measured ungly, maybe due to the "non-optimum" layout.
I can't understand why that would happen, but it might be interesting to investigate it further, as I think that the problems are more likely to be ones that can be solved than with the other approach.
> So I put the inner shield back to GND with reasonable results - on the bench.
But I presume with zero bias....
> Yes, the bias drives the common and the supplies of the floating DCDC used > for all the OPAs related to the TIA, to bee seen at p. 13 of the document: > 20210316_SuperSIMS_FAMEIO_MV40-Ersatz_Praesentation_Kirschke.pdf. > > The U7-A at p10 is an INA149, a TI HV-Diffamp. I like to simplify the structure a bit, > omitting the composite amp. It's power is GND-referenced, of course. > But it's input draws already ~0,5mA from bias and ~0.5mA from the TIA's output. > Because the bias current possibly available at any in/out of the module must be > limited to remain <3mA under all conditions for safety purposes, > all power sinks should be kept at minimum.
It might be interesting to float the ADC and just transfer the digital result from the vbias-referenced ADC to the rest of the system with an optocoupler or one of the newer equivalent transformer or capacitive coupler devices. Especially since you are not going to get super-high accuracy with teraohm resistors, the ADC and voltage reference might as well be cheap ones (even the built-in ones of a microcontroller) so there is less incentive to share one expensive reference + ADC amongst multiple channels.
>> ...replacing the switchable feedback resistors by a p-n junction... > > Our "customer" physicists asked for an idea of such a version, so I simulated one, see > 20210522_FAMEIO_SuperSIMS-MV40-Ersatz_S45-46.pdf > which document was added to the link above. > But it's still not realised by them, will take some additional weeks.
Perhaps you can have a try at home. Sometimes it is faster and if it is promising then you can guide them to replicate it at work.
> For our purposes we'd like to get some more reliable accuracy at least at the higher ranges. > We bought the resisors at SRT-Restech, the 1T in 2512 has 5% TK250, you're right.
Thanks for that info. I did not even really know where to get ones that good, I think mine are 30% or worse, even before I take them out of the packet.
> But the range _could_ be calibrated digitally, if necessary. > But the log-idea itself is really good, i believe. > On the other hand I do not have any practical experience with logamps...
Nor do I. I think the main challenges would be bandwidth variation and getting the compensation right for all current levels, as well as getting good accuracy in spite of temperature gradients and poor log conformance at high currents. If the input current signal has a high peak-to-average ratio and higher bandwidth than the circuit, then such a scheme might also suffer from the same problem as older spectrum analysers where one can measure "the average of the log of the signal" which is not the same as "the log of the average of the signal" that one would prefer to measure.
On Monday, May 24, 2021 at 10:23:28 PM UTC+10, Chris Jones wrote:
> On 24/05/2021 21:14, timo.k...@ibtk.de wrote:
<snip>
> > But the log-idea itself is really good, i believe. > > On the other hand I do not have any practical experience with logamps...
I've used them, and they do work. Bob Pease has published quite a bit on the practical problems. Setting up a circuit where an op amp makes the voltage at the collector is the same as the at the base can give you good logarithmic behavior over about seven orders of magnitude. "The Art of Electronics" makes the point in fig 2.41 in the third edition, 2.39 in the second edition.
> Nor do I. I think the main challenges would be bandwidth variation and getting the compensation right for all current levels, as well as getting good accuracy in spite of temperature gradients and poor log conformance at high currents.
If you use one half of a monolithic pair of transistors as your logging device, temperature compensation can be pretty accurate. The voltage drop in the base resistance can be a problem at higher currents, but you can compensate it out if you monitor the base current and correct for the drop - I worked with hyperboloc function generator which did that.
> If the input current signal has a high peak-to-average ratio and higher bandwidth than the circuit, then such a scheme might also suffer from the same problem as older spectrum analysers where one can measure "the average of the log of the signal" which is not the same as "the log of the average of the signal" that one would prefer to measure.
That's always a problem with non-linear circuits. -- Bill Sloman, Sydney
Chris Jones schrieb am Montag, 24. Mai 2021 um 14:23:28 UTC+2:
Hello Chris,
> Sorry I didn't have time to read your documents...
You couldn't do that, because I added some afterwards. ;)
> I have to do something like this, I would use two cascaded toroidal > transformers...
That sounds very interesting. Are you sure, that toroidals are best-suited with respect to their stray field? Years ago I read anywhere, they're not, I believe from Bruno Putzeys. So the best classD-Amp modules (as far as I know) have RM-cores instead of toroidals. See here (lower left): https://ibtk.de/project/amplifier/Power-Amp/Fotos/20191004_Power-Amp_4xNC252MP_open_IMG_2149_1k.jpg The cap coupling could be worse, don't know...
> I would not > expect it to be useful below nanoampere signal levels unless I could > arrange for the transformer operating frequency to be so far above the > frequency of any wanted signals that it could be filtered out very well.
I'm lucky to have just this case. The (higher) currents to be measured are well below 10kHz, the rest will be 4th order lowpassed before the ADC, the (sine driven) transformer will be ~100kHz or above.
> > So I put the inner shield back to GND with reasonable results - on the bench. > But I presume with zero bias....
For a short period with bias too. Very noisy, but seemed to be stable. And when I switched off the bias, it's 1uF buffer cap put the charge into the +In of the ADA4530-1. One shot only. Expensive and sometimes hard to get... The layout is a little bit complex, the Rogers is sensitive to multiple times soldering, there are some more Guard-related mistakes, so went back to design a more robust version, hopefully.
> It might be interesting to float the ADC and just transfer...
It would throw out the INA at least, worth to think about, thanks. Maybe I find the chance to build up a log version too... I added the offer list from SRT to the small website, if you are interested. Thank you! Cheers, Timo
Hello,

Thank you Bill, for your hints!
I will study the AoE chapters.
So it seems to be worth an practical try.

Cheers, Timo
On Tuesday, May 25, 2021 at 12:10:50 AM UTC+10, timo.k...@ibtk.de wrote:
> Chris Jones schrieb am Montag, 24. Mai 2021 um 14:23:28 UTC+2: > Hello Chris, > > Sorry I didn't have time to read your documents... > > You couldn't do that, because I added some afterwards. ;) > > I have to do something like this, I would use two cascaded toroidal > > transformers... > > That sounds very interesting. > > Are you sure, that toroidals are best-suited with respect to their stray field? Years ago I read anywhere, they're not, I believe from Bruno Putzeys.
Toroidal transformers have no stray field, if they are non-progressively wound. If you just wind wire around the toroid, you end up with a turn in the plane of the toroid, and you have to organise the winding to avoid this. One scheme is to wind clockwise half-way around the toroid, then wind back anticlockwise all the way around the toroid - eventually going over the top of the first winding - and then wind back anticlockwise over the top of the second layer until you get to your original starting point. I first found out about this by reading https://www.amazon.com/Coaxial-AC-Bridges-B-Kibble/dp/0852743890 The Kibble involved is the one remembered in the Kibble Balance. Later, I looked at using a pair stacked toroids to measure the conductivity of a fluid, and that only works if both coils are non-progressively wound.
> So the best classD-Amp modules (as far as I know) have RM-cores instead of toroidals. See here (lower left): > https://ibtk.de/project/amplifier/Power-Amp/Fotos/20191004_Power-Amp_4xNC252MP_open_IMG_2149_1k.jpg
Toroids are a pig to wind. There are special machines for the job, but RM core parts are a lot easier to put together in production. Pot cores (and RM cores are particular sort of pot core ) don't have much stray field. Topologically they are the inverse of the toroid - you wrap the core material around the winding, rather than the winding around the core - but the slots to let the wires out means that they aren't as perfect as a toroid. <snip> -- Bill Sloman, Sydney