Precision high-voltage pulse settling-time measurement

What are your actual measurement goals?

I would expect that diode clamp to have all sorts of parasitics that don't show up in simulation.

I had a horrible time with the scope monitor pickoff in my 1200 volt,

Unequal diode-chain currents, and worse. Don't trust your SPICE simulation, poor models, and uncertain parasitics.

Make a simple low-distortion capacitive divider. 1:25, get the voltage down to the 10-volt region where you can deal with it. Maybe clamp the resistive output of the divider with Schottky diodes.

If James has a sampling scope, he only needs to divide down far enough for the bridge to be safe. Those things are the bee's knees for settling measurements because you don't peg the front end, so there's no significant settling tail in the scope itself.

Cheers

Phil Hobbs

So why not a resistor divider pair? George H.

Or if it's fast a capacitor divider chain? GH

It's hard to state hard goals because the client simply always asks for the best I can possibly do. But, the driving force is a need to make sure the pulse *really* goes to zero volts as quickly as possible. I infer that a few hundred mV ain't zero, that's bad.

The operator has to tweak the load for best response, and I'm figuring they'll need an honest 'scope pickoff to know when they've succeeded.

Probably. The construction's strays alone, for one. But I did think putting the diode capacitances in series was fun.

I'll probably use the classic 'scope divider in real life. I erred by splitting the divider resistor in a 'scope monitor pickoff once. I did it to get a higher voltage rating, but stray capacitance at the join made a ghost settling tail, and a bit of tail-chasing to find it.

Cheers, James

Clamping is a 'maybe', I've considered that, but I'll have to do some figuring. The divided node is going to be hi-Z, and even 1pF stray on a 10k node (or 100k) would be ugly.

A conventional divider is almost certainly going to be ultimately necessary though, since I'm charged with scaling this up to 3kV for another effort.

Cheers, James Arthur

I want to watch the settling tail in detail that a divider would obscure, e.g., 250V->10mV in

A shorted delay line returns a pulse to zero pretty well (lots of literature on this, related to proportional counters). Look up 'delay line amplifier'.

I've got a 7S14 in mothballs, which, with a new sampling battery, would do. But effectively including the sampler front-end in the product could be a convenience to the users, that's what I'm day-dreaming about.

It's possible to non-electronically gauge the settling performance by way of examining the load instrument's results. That's sensitive, but rife with ambiguities and massively inconvenient.

Cheers, James

It's an impedance-matching problem. The pulser is hellaciously fast and clean at its output, but the load sends back reflections when the output's impedance isn't perfectly matched. That's what they're tweaking, guided by 'scope pickoff observations.

Cheers, James Arthur

A quick google search turned up this:

Have you even tried to do your own search?

No, I didn't. The signal-to-noise ratio was too unlikely to be worth it.

Your reference could be great for a newbie, for example, but it's three orders of magnitude too slow, with an offset voltage that's an order of magnitude larger than the signal I'm measuring.

I'm well familiar with conventional techniques.

I did review some old Jim Williams' app notes on instrumenting fast op-amp settling times; those are always fun.

Cheers, James Arthur

I have this design for a $2 single-diode sampler with 150-ps resolution that I did with CW a few years ago. Haven't measured its sampling artifacts down at the 100 ppm level, but it seems to work fine even without a sampling loop.

(In single-diode samplers both ends of the sampling cap bounce all over the place, so making a sampling loop is a bit of a puzzle.)

It would be fun to try putting it in something else. I should do that one of these times.

Cheers

Phil Hobbs

Spherical cows rule! (Or is that 'roll'?)

Another approach would be to clamp the capacitive divider using a pHEMT with a sliding gate pulse. You can turn those off in way under 100 ps if your gate drive is quick enough, and the newer 18-GHz ones are much faster than that. The drain-gate capacitance is very small.

JL says they make good diodes, so you could protect it by coupling the pulse itself into the gate as well as the drain.

There would be a bit of DC offset, but nothing too hard to calibrate out.

Cheers

Phil Hobbs

My noodling on this started with a conventional sampler:

.------------. | | .--+--. | | | | V V | --- --- | | | | Vi >----+ +------------> Vo | | | V V | --- --- | | | | '--+--' | | | --. .-' | )||( | )||( | )||( | )||( | --' '--------------'

Then I thought, heck, if I had a suitable 250V schottky I wouldn't even need to sample -- I could chop off the high-voltage portion entirely and go continuous-time.

Like, if we strapped some wings on a donkey, maybe it could fly. :)

Cheers, James Arthur