HPM7177 ADC from CERN

[View[+]Finder]

The AD5791 demo board, that will be the reference for my AD7177 imitation of CERN's HPM7177 (see earlier posts), now has the LTZ1000 add-on board. I've been running it 24/7 for over a month and will make the results available here. As is the case with the HPM7177, my expected result will be a DC voltmeter that is very good at one voltage range, not a DMM by any measure. What I hope to get is a combination of a better precision reference and a means of verifying other meters in a limited way.

[Echo88]

@Castorp:
I have a question regarding the power supply section.

Why was the decision made to use a medical grade AC/DC-brick (TPP 30-112-D) instead of a power supply as used in the PCB/Fluke 7001, when CM-noise is a concern?
Too much hassle since the big transformer (since it also powers the TEC) and its conductive shell would be a custom part?

https://edms.cern.ch/ui/file/2606855/1/EDA-04259-V2-0_sch.pdf
https://eprintspublications.npl.co.uk/1888/1/bemc99-7.pdf

[Echo88]

I intend to replicate the HPM7177 ADC board, since its a very good and tested basis for contemporary ADC-experiments.
For my purposes i`ll modify it a bit:
-different pcb size/format
-4 Layers instead of 6 if it doesnt degrade the specs, but one gotta say the 6 Layer stackup is very nice
-TDP-networks instead PRND, dont want to wait half a year and spent more for them than the LTZs

The attached document shows the usual HPM7177-schematic, modified with TDP-networks and comments.
Also the last two pages show ideas on how to modify it for high impedance input by using JFET-OPs or discrete JFET-implementations.

Hope for some fruitful discussion. The only HPM7177-copy that i know off is the one from Marco Reps, which he sold.

[Kleinstein]

The resistors for the reference scaling and gain setting for the ADC input are important for the performance. With lower grade resistors one may have to do a slightly more frequent check of the gain by measureing the raw 7 V reference.

The version with the discrete JFETs would normally not need an extra buffer between the input buffer and the differential amplifiers at the ADC inputs.

The extra JFET to drive the bootstrapped drain voltage could be a weak point: the JFET is behind the loop and drain-source voltage (and thus the common mode voltage) effects the gate source voltage.
This may be good enough for the 34401 but may give INL errors with higher demand. It may help to trim the get currents, to get a low drift for the FET pair.

Modern JFET OPs (e.g. OPA140) are quite good and could also be used with a bootstrapped supply. So there is limited need for the discrete JFETs.

The input switching part does not provide a path for a precharge phase. So switching at the input can cause quite some current spikes. Nomally one should have an extra high impedance buffered signal as an extra input path to reduce the spike for critical inputs. Depending on the use, there is limited need for the +-10 V ref. levels. The scale factor calibration would use the +7 V reference and ground only. The extra switch chip before the MUX is not really useful. The main mux should already have ground as one input - if more inputs are needed, have an extra MUX for some reference levels like the 5 V or VCM that are less critical. The 7 V ref to the MUX chips should have a series resistor to limit current spikes to the reference.

The circuit does not provide protection for the input - this would likely be a part to add.
The is usually no 10 Gohm resistor at the DMM input - the specs are usually something like > 10 Gohm  and here the > can also allow for a lot larger.

[Echo88]

Input protection was omitted for clarity in the schematic, in the end its needed of course.
I was unsure wether the 10G-resistors were needed at all for these low biascurrents of the OPA140, just included them as a biascurrent-path, i guess they could be made bigger.
The 34401A JFET-frontend is quite nice and simple, but it most likely will not be good enough for the best contemporary ADCs.
One could use the 34420A-frontend, but the selected IF3602 is a difficult point.

You raise an interesting point with the precharge phase and questioning the many mux-voltages. I need to read up about it.

With 140dB CMRR the OPA140 is already quite good for this task, but bootstrapping is a nice exercise in general.

[branadic]

One could use the 34420A-frontend, but the selected IF3602 is a difficult point.

Are you sure the IF3602 were selected, as it already is a matched pair JFET, not the single JFET version IF3601.
If you don't mind you can select yourself from 2SK147 to create yourself a 2SK146. There are similar JFET devices such as 2SK369 or IFN147/IFN146, LSK389, 2SK117, 2SK170 still available today.

-branadic-

[Kleinstein]

With only a buffer and than a gain off less than 1 towards the ADC there is no need for super low noise FETs or amplifiers. One should get away with relatively normal FETs (e.g. SK209, SK2145, 2N4393).
The 34420 uses the special FETs to get super low noise for the cases with high gain, not for very good linearity.

The 34401 type circuit uses the FETs as source followers and this is still limited by the linearity of the following OP-amp. Meters like the 34420 or 3458 use the FETs already as a first gain stage and this way  add to the loop gain and the CMRR of the OP-amp is not critical.

Also a good JFET OP-amp is a very viable option. Bootstrapping the supply can help to improve the linearity. One still has the output cross over distortion, that may need class A biasing.
With bootstrapping the circuit is however not that much simpler than the discrete JFET version. Discrete FETs can reach lower noise, but with drift OP-amps are usually easier.
Also a zero drift OP-amp can work, though it needs some compromising between noise and bias current. With there usually very good CMRR and loop gain one might get away without bootstrapping the supply, though there are more types available for a 5 V supply.
 
The high impedance DMM inputs usually have no intentional resistor to ground. So the open circuit behaviour is undefined and the bias current may as well make it drift all the ways to saturation. This is not a problem and the DUT will provide the DC path.

There is nothing wrong with many possible signals to the ADC. Chances are there will be even more with precharge. One can separate the less critical ones to a 2nd MUX in a side path so leakage there would not effect the sensitive inputs that only go through one MUX chip.

For the leakage and bias the actual performance can be quite a bit better than the maximum specs and it may be hard to gurantee low leakage from the spec limits. The actual performance can be quite a bit better. It could be viable to build based on typical specs and still get a good yield (low chance to find excessive leakage) for the input section.

[Echo88]

Apologies, the pdf in post #152 is wrong on the 5. page: its missing the OPA140-section and instead is just a copy of the discrete JFET-section of page 6.
You are right branadic: i didnt remember correctly. The IF3602-batches arent measured for matching.
According to AoE3 they are most likely measured for their Vgs/Ids characteristics, to ensure the JFET-stage works correctly.

Attached is a new pdf with a simple OPA140-stage on page 5. Also page 7 shows a detailed improved OPA140-based input stage with added features. I hope i implemented the suggestions from Kleinstein correctly.

A JFET-OP-stage is intended since i dont really wanna deal with the current spikes/bias current of AZ-OPs (gonna miss the very high CMRR/AVOL of them though).
The autozero/fullscale-cal done by the Mux together with the temp stabilization by the TEC should be good enough to ensure the necessary stability for this task.

[Kleinstein]

The bootstrapped buffer part still has a few weak points: the voltage range is somewhat limited with just a divider and with only partial (e.g. 3/4) bootstrapping there is only a moderate boost in CMRR.
The in loop buffer to get more driving power should not use the bootstrapped supply, but normal supply: the main idea if this amplifier is to provide the current, as it is hard to provide much power from the somewhat soft bootstraped supply.

The prechearge part is wrong. As shown the extra path works more like a hold circuit, but does not provide an auxiliary signal with the input potential when the input is not yet connected. Instead the extra buffers should be from the inputs directly (before the mux) and the precharge signals should be more going to the main mux as well. The Photomos switches are rather slow to use them for precharge. The buffer for the raw input can also help with the protection (e.g. support clamps).

With a differential input, one would likely need more choices for the input MUX. Currently there is no true differential mode. Chances are one would need independent MUX chips for the positive and negative side. Signals would be something like the pos and neg side input, the pos and neg side precharge signals, maybe ground, the extra MUX for reference signals. One may also want some way to ground (or set to a common mode valze) one of the inputs in case the signal source is not differential, like an isolated input.

The protection still needs the clamping and turn off part. With the photomos or similar as protection, there should be a backup with a PTC and/or fusible resistor.

[macaba]

Some good points from Kleinstein, now adding one of mine:

Consider capacitance of switches
- MUX36D04 is low capacitance mux and the input signals are mostly DC so that should be fine.
- AQW210 in the input mux arrangement might be worthy of simulation, even simulating it as a network of capacitors - the capacitance is much higher (though very low for SSRs in general) so you might have AC signal leakage. Possible way around it - a t-network with 2 SSR ICs, with the middle node connected via high value resistor to the precharge buffer output.

[Castorp]

Nice work, very interesting. Also a good reminder about the large step from digitizer for a specific fixed application to a more general-purpose instrument (regarding the frontend, power supply, etc.)

This year has been extraordinarily busy in many ways, but I finally managed to travel to PTB and do those tests with the PJVS. Now I only have to find time to process all the data 

[Echo88]

Attached is the daily modified version, last page is of interest.
Only the positive input is complete, the negative one neglected as the schematic is overthrown every day.
Also lets view it as a +-10V voltmeter, just this range and just sufficient protection to survive lets say 50V.
You use Eagle, right Kleinstein and i assume Macaba uses Altium?
Otherwise i could share the kicad-project file to those interested in participating, makes it easier than do the schematic from assuming what you meant and Kleinstein giving up on me. 
The 34420A and 3458A frontends are interesting for this task naturally, way more JFETs (and their BOOT-bias...) instead of Muxes.
 
Come on hiding greybeards, show us how its done nowadays. 

[Kleinstein]

For looking at the schematics a PDF file is OK, though some would prefer a PNG.
Kicad files are a bit tricky to share with many references to part libs and different lib versions and paths.
For reference it helps to have the parts numbered

The switching still looks overly complicated. For a single input (other side to ground) the critical input should to go to few switches. In the minimal solution there are just 2 switches to contribute to the leakage:
1 from the input to the the main amplifier and
1 from the amplifier input to a 2nd MUX with all the less critical signals (references, ground, precharge signal,...).

In the old times the CMOS switches were limited and JFETs gave better performance. Modern CMOS switches got quite good and are easier to use. If a suitable gate signal is avaible JFETs are still a viable solution.  JFET solutions may have quite some charge injection. CMOS switches are often quite good in compensating the charge spikes. However the charge injection is usually voltage dependent and the datasheets often give one of the better operating point.

The topic of a good DMM input is quite lengthy. So it may be better so open a new thread for this, to keep this thread more focussed on the HPM7177.