High voltage, high speed measurement

[Optikon]

I'm brainstroming topologies for a circuit that can perform the following:

1) Maximum differential signal can be as high as 300V, either polarity. (300 peak) AC or DC

2) For AC measuring, bandwidth required is at least to 100kHz, faster would be even better. Needs to be "good" beyond 100kHz to get a clean 100kHz measurement (low distortion, say <0.1%)

3) Measurement circuit must not load input signal. I.e., high impedance, how high? not sure yet but you get the idea.. the higher the better over frequency.

4) output will get scaled to a 10V peak (FS). 300V in = 10v out.

5) all the usual nice things like good linearity, low drift etc... I'd be willing to compromise on some of this though if a tougher problem is solved like the relatively high bandwidth.

Well, I think this is a tough nut to crack. I've considered high voltage input instrumentation amps and my own discrete versions of the same. I'm not too happy with this idea mainly because of the god-awful bandwidths of such contraptions..

Now I'm considering a discrete transistor high voltage front end stage. (I think High voltage mosfets are a must and biasing them is a drag.)

Who's got ideas? While I try and work this out, I'll entertain any thought fragment, or block diagram or comments. Much appreciated!

Oh, one more thing, this is a from scratch design so I do not have restrictions on things like power supplies available etc.. (although, I dont wanna have to make a 300v+ supply - thats ugly in many regards) and I'll consider doing whatever might work..

Thanks!

 

[Styx]

voltage divider to pot the signal down (good quality resistors)
use a isolation amplifier to get a differential voltage signal

 

[zachtheterrible]

i think you can find high voltage probes for o-scopes?

 

[Roff]

I'm not sure I understand. Is the input really differential (two input ports), or is it a +/- 300 volt peak signal relative to GND (600V p-p), or what?

 

[Optikon]

In my haste, maybe input requirements not so clear.. here is a pic.

 

[Optikon]

voltage divider to pot the signal down (good quality resistors)
use a isolation amplifier to get a differential voltage signal

Yes, I thought of trying this. If I just use resistors, I don't think I can get to the bandwidth I need (the resistors will be large & LPF with parasitics)

However, I might be able to make impedance networks that are resistive only at DC. But the prospect of matching capacitors etc.. seems unappealing.

 

[Optikon]

i think you can find high voltage probes for o-scopes?

Thanks for the tip Zach. However, I am trying to design a high voltage measurement circuit. I may be able to borrow the ideas of these probes though. I think they might be resistors & differential FET amplifiers? Not sure..

 

[Nigel Goodwin]

i think you can find high voltage probes for o-scopes?

Thanks for the tip Zach. However, I am trying to design a high voltage measurement circuit. I may be able to borrow the ideas of these probes though. I think they might be resistors & differential FET amplifiers? Not sure..

A x10 probe is just a 9Mohm resistor, with a small trimmer capacitor across it - it forms a potential divider with the 1Mohm impedance of the scope and it's input capacitance.

 

[Optikon]

i think you can find high voltage probes for o-scopes?

Thanks for the tip Zach. However, I am trying to design a high voltage measurement circuit. I may be able to borrow the ideas of these probes though. I think they might be resistors & differential FET amplifiers? Not sure..

A x10 probe is just a 9Mohm resistor, with a small trimmer capacitor across it - it forms a potential divider with the 1Mohm impedance of the scope and it's input capacitance.

I am aware of this... this is the usual passive 10x probe. I have a set of matched differential "high voltage" probes that provide 10x or 100x attentuation and the enclosure they connect to looks way too large to be just some resistors and caps.. also the power supply wire to it seems like a giveaway that it has an active circuit in it... not sure if it's what I need.

I also like the idea of active probes that can provide G-Ohms of input impedance for light loading.

But, if I use a regular divider, I'll need two tuning capacitors(one for each side) and they'll need to match and stay put over time.. (asking for a tweak from time to time will be unacceptable) not sure if I'll have trouble with such a simple arrangement.. though I like the simplicity of it!

Any one else with ideas? Anyone think I can get away with a high voltage discrete diff amp for input?

Thank you all so far!

 

[Roff]

The possibility of transformer coupling on the input implies that you don't need response down to DC. Is that correct?
If you can use a transformer, you can use a step-down transformer, which will preserve your bandwidth, give you the attenuation you need, and make it easy to have a high input impedance , since Zin equals Zload times the square of the turns ratio.

 

[Optikon]

The possibility of transformer coupling on the input implies that you don't need response down to DC. Is that correct?
If you can use a transformer, you can use a step-down transformer, which will preserve your bandwidth, give you the attenuation you need, and make it easy to have a high input impedance , since Zin equals Zload times the square of the turns ratio.

Input could be AC or DC.. I only mention transformer coupled input to illustrate that input circuit might not be referenced to ground. Though, connecting input side to ground and applying 300VDC is still a required case.

So far my two leading ideas are:

1) Go with resistor dividers with a "flatening network" to keep frequency resonse flat up to requirement. follow this with a sufficiently high speed diffamp. This works down to DC.

2) Try to build a high voltage FET input diffamp that can simply handle the 300V requirement directly. Flattening network might still be needed but input impedance would be much better than other idea. Biasing, making low drift & linearity might be difficult.

Any other thoughts to throw out?

Thank you all so far for the great input!

 

[Roff]

It would really be helpful if you could define the maximum load you can tolerate. I don't see how you can proceed without defining it.

 

[Optikon]

It would really be helpful if you could define the maximum load you can tolerate. I don't see how you can proceed without defining it.

Ron, my input stage will connect to a variety of voltage sources producing various transient effects (up to 300V peak) and are desirable to measure (accurately in a 100KHz bandwidth). Some of these sources will not have a low output impedance (how high? I dont know how to characterize that and it wont be constant anyways.) So for starters I will say that an input impedance (up to 100KHz) of less than 10 MegOhms will be unacceptable.

Right now a solution for these kinds of measurements is with an oscilloscope so my input requirements are very similar but I also desire an improvement in loading effects if it can be done. I will not be digitizing waveforms like a DSO does. Accuracy needs to be better than 0.1%. If better is achievable, then that will be determined by cost constraints..

I'll admit, alot of the detail is not worked out yet, but I'd like to throw away dead-end ideas as soon as possible.. but I've only got two ideas that feel like I can get there.

Any help = much appreciated!

 

[Nigel Goodwin]

So far my two leading ideas are:

1) Go with resistor dividers with a "flatening network" to keep frequency resonse flat up to requirement. follow this with a sufficiently high speed diffamp. This works down to DC.

That seems the obvious option, which was why I mentioned a scope probe, use the same technique to compensate your input attenuators. As you're wanting high frequency (although 100KHz isn't particularly high), you will probably need to go the 'scope probe' type route, simply due to the capacitance of the connecting cable (which is why the resistor is at the probe end).

2) Try to build a high voltage FET input diffamp that can simply handle the 300V requirement directly. Flattening network might still be needed but input impedance would be much better than other idea. Biasing, making low drift & linearity might be difficult.

Sounds impractical?, and pretty pointless?. You're not wanting to amplify anything, just the opposite, you want to attenuate something considerably. Blasting 300V directly into a FET isn't going to reduce the signal much?, and the amps can only make it larger?.

Input impedance isn't a problem, with 300V signal level, keeping to the standard scope type impedance of 1Mohm you could use a x100 probe (or make a similar probe), giving you 100Mohm input impedance, and 3V output from your buffer amp.

 

[Oznog]

We could help you more if you could describe in a bit more detail what you need to do.

If you're building a meter that can be attached to an external source, the scope probe might be the way to go.
If it's a permanently wired source you'd likely just build your own divider.
If you just want to read a signal on the bench, you'd preferrably just use a bench scope!

My scope has a probe rated for 600v. The scope is only rated for 300v, but it's a 10:1 probe so a 600v signal only provides 60v to the source. Be aware of the difference between an AC RMS voltage rating and a peak DC value rating.

Most line-powered scopes tie the probe ground to the scope chassis and the power line earth ground. As such you can never put the scope probe ground on any wire which is either not tied to ground or completely isolated from AC ground. But battery-powered scopes are nicely free from this problem, and they use all rubber and plastic buttons and case to provide good personal protection to keep the user from contacting the potential which may be present on scope ground.

It is not necessary, or is it practical, to build an amp for 300v. Most op amps are in the range of 5v-15v. You just a divider with 2 resistors to reduce the signal to the desired level.

 

[Roff]

Here's a diff amp that simulates nicely. It has excellent CMRR. You might get away without the compensation caps (Cc) in the real world. Obviously, you need precision resistors and/or several pots, but this is true of any diff amp. You don't need caps across your 100Meg resistors, which is a big plus, I think.
U3 may not be necessary, but it lowers the CM gain on each input amp, possibly making balance in the output diff amp less critical.

 

[Optikon]

Here's a diff amp that simulates nicely. It has excellent CMRR. You might get away without the compensation caps (Cc) in the real world. Obviously, you need precision resistors and/or several pots, but this is true of any diff amp. You don't need caps across your 100Meg resistors, which is a big plus, I think.
U3 may not be necessary, but it lowers the CM gain on each input amp, possibly making balance in the output diff amp less critical.

Ron thank you for the design reference!

This looks like an instument amp? What is the CMRR of this particular one by the way? I am thinking I might try an integrated INAMP with built in, ratio matched resistors for maintaining CMRR. What do you think about that? I have bandwidth worries on the integrated version though - I'll have to see. If I go discrete, I'll need precision networks I believe to keep CMRR in check. Can you further explain how U3 lowers common mode gain on the input amps?

I wont need any pots. A system cal will exist to eliminate offsets & gain adjustment. (there is alot more to this design that is outside the scope of this isolated piece) I do need this part to be low drift. It needs to operate within spec having a 30 deg C ampbient temp change. I think it's do-able with approriate choice of opamps & resistors. Low drift 100Meggers though might be $$$.

Thanks!!!

 

[Roff]

Here's a diff amp that simulates nicely. It has excellent CMRR. You might get away without the compensation caps (Cc) in the real world. Obviously, you need precision resistors and/or several pots, but this is true of any diff amp. You don't need caps across your 100Meg resistors, which is a big plus, I think.
U3 may not be necessary, but it lowers the CM gain on each input amp, possibly making balance in the output diff amp less critical.

Ron thank you for the design reference!

This looks like an instument amp? What is the CMRR of this particular one by the way? I am thinking I might try an integrated INAMP with built in, ratio matched resistors for maintaining CMRR. What do you think about that? I have bandwidth worries on the integrated version though - I'll have to see. If I go discrete, I'll need precision networks I believe to keep CMRR in check. Can you further explain how U3 lowers common mode gain on the input amps?

I wont need any pots. A system cal will exist to eliminate offsets & gain adjustment. (there is alot more to this design that is outside the scope of this isolated piece) I do need this part to be low drift. It needs to operate within spec having a 30 deg C ampbient temp change. I think it's do-able with approriate choice of opamps & resistors. Low drift 100Meggers though might be $$$.

Thanks!!!

I don't think you can get inamps with inverting inputs. With non-inverting inputs, you will need passive attenuators and scope-probe type compensation. The idea in my design is to convert the input voltage to a current (into virtual ground), thereby avoiding the caps.

Regarding U3: The differential signals on U5 and U6 outputs cancel each other at the inverting input to U3, but any common mode signal on those outputs gets compared with ground (non-inverting input) on U3, amplified by a zillion (or is that a brazilian?) and fed back to the inputs of U5 and U6, forcing the common-mode signals on their outputs to (nearly) zero.
If the input stages are perfectly matched, the change in CMRR at the output is negligible, but the CM signals at the outputs of U5 and U6 go down dramatically with U3 in place. This improves your dynamic range in the presence of large CM signals, and reduces possible IM distortion between Vcm and Vsignal.
Below is a plot of CMR. The CMRR is 40 dB worse than this, because the diff. mode gain is -40dB. Still not bad. remember this is with perfectly matched components. I encourage you to simulate it, and if you like it, build a breadboard.
If you don't like 100Meg resistors, you could use 10Meg if that is high enough, and eliminate R3, R4, R7, and R8. R3=R7=0, R4=R8=∞.

 

[Optikon]

Here's a diff amp that simulates nicely. It has excellent CMRR. You might get away without the compensation caps (Cc) in the real world. Obviously, you need precision resistors and/or several pots, but this is true of any diff amp. You don't need caps across your 100Meg resistors, which is a big plus, I think.
U3 may not be necessary, but it lowers the CM gain on each input amp, possibly making balance in the output diff amp less critical.

Ron thank you for the design reference!

This looks like an instument amp? What is the CMRR of this particular one by the way? I am thinking I might try an integrated INAMP with built in, ratio matched resistors for maintaining CMRR. What do you think about that? I have bandwidth worries on the integrated version though - I'll have to see. If I go discrete, I'll need precision networks I believe to keep CMRR in check. Can you further explain how U3 lowers common mode gain on the input amps?

I wont need any pots. A system cal will exist to eliminate offsets & gain adjustment. (there is alot more to this design that is outside the scope of this isolated piece) I do need this part to be low drift. It needs to operate within spec having a 30 deg C ampbient temp change. I think it's do-able with approriate choice of opamps & resistors. Low drift 100Meggers though might be $$$.

Thanks!!!

I don't think you can get inamps with inverting inputs. With non-inverting inputs, you will need passive attenuators and scope-probe type compensation. The idea in my design is to convert the input voltage to a current (into virtual ground), thereby avoiding the caps.

Regarding U3: The differential signals on U5 and U6 outputs cancel each other at the inverting input to U3, but any common mode signal on those outputs gets compared with ground (non-inverting input) on U3, amplified by a zillion (or is that a brazilian?) and fed back to the inputs of U5 and U6, forcing the common-mode signals on their outputs to (nearly) zero.
If the input stages are perfectly matched, the change in CMRR at the output is negligible, but the CM signals at the outputs of U5 and U6 go down dramatically with U3 in place. This improves your dynamic range in the presence of large CM signals, and reduces possible IM distortion between Vcm and Vsignal.
Below is a plot of CMR. The CMRR is 40 dB worse than this, because the diff. mode gain is -40dB. Still not bad. remember this is with perfectly matched components. I encourage you to simulate it, and if you like it, build a breadboard.
If you don't like 100Meg resistors, you could use 10Meg if that is high enough, and eliminate R3, R4, R7, and R8. R3=R7=0, R4=R8=∞.

Thanks Ron, I'll explore this topology further.