Input protection recommendation for ADA4255?

[dobsonr741]

I'm about to build a DAQ experimentation board with ADA4255 for hobby use. The part is nice, has protection up to +/-60V already built in.

I'm about to operate it from isolated power and SPI, so common mode voltage is not a major concern. I'd like to extend the input protection, so that it can survive 500V across the inputs.

It will auto range by it's own internal PGA, 20V being the highest direct voltage range, and I am to add a 200V range with a 10M 10:1 divider via a relay.

What topology would you recommend for the input protection of the two differential inputs? Ground would not be used. 

https://www.analog.com/media/en/technical-documentation/data-sheets/ada4255.pdf

[dobsonr741]

The answer is always "Datron". What do you think of this?

[dobsonr741]

On closer inspection, realized I’ll need to bootstrap at least the antiparallel JFETs, to have meaningful results.

[Terry Bites]

polyfuse +TVS?

[dobsonr741]

This is what I came up with. MOSFET current limiter instead of polyfuse, bootstrapped zeners (could be a TVS). It should survive 500V, while accurate to 6.5 digit from 0.2V to 200V range and do not degrade the otherwise very high input impedance of the ADA4255 up to the 20V range. 

[Kleinstein]

The OP-amps for bootstrapping the zeners should be more like TL032 or TLV9302 low bias OP-amps and not zero drift ones because of the switching spikes.

For the divider is really depends on the application. Even if the meter part has a differential input, one still needs to provide a common mode level and has common mode coupling from the supply.
So there needs to be link to ground too. This could be an extra guard terminal.
Essentially all the DMMs have one terminal tied to ground or a low impedance node of the input section. So the 2 inputs are not symmetric and the divider is 1 sided, more like the earlier version. One terminal can be low impedance and get away without protection.

Already the 100 K resistor in the earlier version could be good vor 500 V, provided the resistor can withstand the voltage. The downside is a little extra noise, though possibly an acceptable amount.
The protection with depl. FETs allows less noise, but it may fail in an unsafe way with a short. So one should have some secondary protection with something like a polyfuse / PTC or fusible resistor to limit the damage after a fault. The polyfuses often also have a rather limited voltage rating, but there are some higher voltage rated PTCs intended for DMMs (especially the ohms ranges).

For ESD protction one may want an extra spark gap + MOV, as the voltage rating of the FETs is limited.

[moffy]

The MCP6V52 goes from input bias current of typ: @25C +/-80pA to @125C +/-1.4nA that would reduce your resolution on your 10:1 divider scale.

[dobsonr741]

The OPA stands out as the most challenging among the options. ADA4255 offers 500uA supply off it's charge pump per rail at +/-22.5V, while MCP6V closely matches but 2 channels will exceed the current budget.

I'm searching for other components that can operate at these high rail voltage levels with a current of <500uA for one channel, after considering @kleinstein's asymmetric conversion.

Could you recommend any OpAmp that fit this description? Let's find the perfect component for this application.

[Kleinstein]

OP-amps for more than 40 V (+-22 V) are rare and than not many low power ones. The OPA990 at least work up to 40 V (42 V max) and is relatively low power (120 µA). Worst case one may have to add dropers (e.g. LEDs / zeners) at both ends.

As bootstrapping the zeners does not have to be supper accurate, one may also consider a pair of mmbf4117 discrete JFETs selected for sufficient breake down voltage.  AFAIR the typical breake down ins in the 70 V range.
The ADA4255 anyway has relatively high input bias and on it's own could be an issue for the divider.
The main nice point is not the input part, but the rather stable gain steps. The linearity is already not really good for the 6 digit level and the input specs of > 1 Gohm can also be an issue with a 1:10 divider with 1 M at the lower end.

[dobsonr741]

The high bias might call for buffering the input then. Would you still consider the auto zero MCP6V5 be a noise concern in a buffer role?

[Kleinstein]

The MCP6V52 may be acceptable as a main input buffer, though in many aspsect the LTC2057 is the slightly better (but close in many points) amplifier.
This may than be instead of the ADA4522, not necessary as additional buffers at the input.

A big question is what is the circuit used for and what are the required properties ?

[T3sl4co1l]

Hmm, should probably be three zener pairs, if you want to clamp Vdiff lower then?  Or if you have supplies (adding +/-15V or whatever isn't a show stopper, https://www.eevblog.com/forum/projects/simplest-way-to-get-12v-or-15v-from-3-3v-but-at-low-power/msg5155689/#msg5155689 ), the clamp diodes can be bootstrapped similarly.  Bootstrapped zeners though, presumably you need one each from line to GND, and then also line-to-line.  This would just be connected where the line-to-line zeners are shown now, no big deal.

What kind of environment is this, what are you filtering/protecting against?  I guess this is an ADC front end, how much precision, bandwidth and CMRR are you after?

Tim

[dobsonr741]

The motivation is to recreate the ADMX3652 - 6½ Digit, ±20 V DVM, then characterize. Optionally, add a 200V range. It seems the 200V support is a completely different ballpark, and creates offset problems I can not address with the 4255's built in AZ. My guess 3652 uses 4255 in the inside.

The protection is just for the bench, to avoid any random mis-wiring or static to destroy the ADA4255 inputs. So this end, the Mosfet limiter, GDT+MOV might be sufficient, no bootstrapping or Zeners, given the fairly good amount of protections already on the Mux input stage.

[David Hess]

Already the 100 K resistor in the earlier version could be good vor 500 V, provided the resistor can withstand the voltage. The downside is a little extra noise, though possibly an acceptable amount.

High frequency noise is reduced by bypassing the series resistance.  Oscilloscope 1 megohm inputs typically use about 470 kilohms in parallel with 1000 picofarads.

However the one example I know of with an oscilloscope 1 megaohm input optimized for low low frequency noise (Tektronix 5A22/7A22/AM502) replaced the series resistance with a 1/16 amp fuse.  My own preference is to use a small high voltage incandescent lamp instead of a low current fuse like some old multimeters did.

[Kleinstein]

The motivation is to recreate the ADMX3652 - 6½ Digit, ±20 V DVM, then characterize. Optionally, add a 200V range. It seems the 200V support is a completely different ballpark, and creates offset problems I can not address with the 4255's built in AZ. My guess 3652 uses 4255 in the inside.

The protection is just for the bench, to avoid any random mis-wiring or static to destroy the ADA4255 inputs. So this end, the Mosfet limiter, GDT+MOV might be sufficient, no bootstrapping or Zeners, given the fairly good amount of protections already on the Mux input stage.

From the bias specs (< 1 nA) and input resistance (> 2 GOhm) it looks like the ADMX3652 is not using the ADA4255, at least not without extra selection.
Chances are the actual leakage is considerably better than the specs and this may be more test limits.
Also the linearity specs of the ADA4255 would need extra care / correction to get the claimed accuracy. Anyway a simple DVM with a ADA4255 front end may still be worth a try, though likely not really high performance. The 200 V range with extrernal divider would get some extra offset, but this could be corrected in software, at least for the fixed part. Alternatively one could have an extra buffer, possibly with a seprate path all the way to the ADC.

If the ADA4255 is used without extra clamps at the input it may be a good idea to have zeners at the charge pumps voltages. So internal clamping to these voltages would not pull these votlages too high / low. With MOSFET current limit a seconday protection ( PTC or polyfuse rated for mains voltage) would be a good idea, just in case. Even a 2 K PTC would not add that much noise or offset from the bias current.

[iMo]

Interesting part the 4255.. +4 and -4ppm gain peaks, but nice symmetrical, easy to compensate in sw, imho..