signal pickoff

How about just a capacitor divider? Use 500:1 area over a common "plane". You can choose the dimensions such that the ratio should be fairly well controlled. It would require a buffer to drive a scope but that shouldn't be too tough.

PG

why such big resistors?

a standard 500:1 scope probe is 10M afaict

why not a string of say 2M resistors and capacitors

-Lasse

You'll want the top resistor to be a chain of smaller ones, or at most, one large power resistor. I'm guessing SMTs would be easier and cheaper though.

All those intermediate pads/traces/bodies will have parasitic capacitance. Shunt that with several pF per resistor. The ratio Cshunt / Cstray determines how much improvement in flatness you get (against slight peaks and dips at middle frequencies, what Tektronix called 'hook', I belive?). For 1/x times better flatness, you need a ratio of, at most, x.

Needless to say, cut out ground plane and surrounding copper, on all layers, around the divider resistors. The characteristic impedance needs to be high. Of course, you won't get it into the megs, but your cap divider will have some impedance at HF. Keep that in mind where the probe part connects to the pulse part: if it has to be a very stable characteristic impedance (say 50 ohms, out into the 100s of MHz), you'll have to account for the probe capacitance too.

The divider then has to have a pretty large capacitor at the bottom, because of the large ratio. Maybe the total probe capacitance is 1pF, so you need

500pF at the bottom. That's a sucky varicap. So don't bother at all. Let the AC gain be what it's going to be, then buffer it, and put a trimpot there. You still need compensation, but it's easier done at DC instead. So instead of a fixed bottom-of-divider R, use a trimpot. Cal process: adjust Rbottom until flat (no leading edge over/undershoot), adjust output gain until AC/DC gain is right.

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
Website: http://seventransistorlabs.com 

"John Larkin"  wrote in message  
news:ouuc4b9o1l2d82samn5v58an6kcdiuc2ut@4ax.com... 
> 
> 
> I'm designing a high-voltage pulser, 1200 volts or so, and I'd like to 
> pick off a divided signal for the customer to monitor. 
> 
> https://dl.dropboxusercontent.com/u/53724080/Circuits/Resistors/Pickoff.JPG 
> 
> Rise/fall times will be a few us maybe. 
> 
> I'd like to have the pulse response be pretty good, so the 
> capacitances matter. The OPA171 opamp and R2 have capacitance, roughly 
> 4 pF total, and I guess the 1G 2010 resistor will, too. These have to 
> be balanced at 500:1 to match the resistors. The cap across R2 can be 
> chosen to work, but I don't want a discrete cap across R1. 
> 
> Maybe the 1G resistor has enough self-capacitance that I can pick the 
> lower cap to work. Or maybe I should add some PCB traces to add 
> capacitance across R1. I need about 0.02 pF if, say, I pad the bottom 
> up to 10 pF. 
> 
> This will be a 4-layer board with L2 ground plane, but I can cut out 
> some plane if that would help. 
> 
> Has anyone done something like this? 
> 
> 
> --  
> 
> John Larkin         Highland Technology, Inc 
> picosecond timing   precision measurement 
> 
> jlarkin att highlandtechnology dott com 
> http://www.highlandtechnology.com 
>

Never, but my first thought on seeing the picture was to try and put the right inductance somewhere. I think in series?

I'd have to do the math.

George H.

What about if the inductance was in series with the 2M? That makes the first bit of current charge up the cap.

GH

Like Tim described, but the intent is to pic a C that swamps out the upper arm stray capacitance. Maybe 200pf to 400pf, and the correct ratio on the lower R. eg; AC divider. I think HP has some appnotes on HV dividers.

Cheers

I have rational reasons for wanting a 1G resistor. Even more would be good, but I can get a 1G 3KV 2010.

It can't be more accurate than your reference, for comparison. HV probes with bandwidth are an issue, the higher the DC impedance selected, but usually just as a result of shielding effects. Without a lot of stray capacitance on the high end of the divider, it becomes easier.

SOT caps on the shunt, may end up being fairly predictable. Matching tempcos and aging of Cs and Rs in the divider can be an issue. What kind of calibration procedure is intended?

The source would need to be ~ insensitive to a double probe load. The effect of removal of either can be witnessed.

If just a monitor, perhaps precision isn't so important?

RL

f.JPG

Wouldn't a trans impedance amp be faster? Then you wouldn't have to move th e summing node. You might have to make an ultra-low capacitance buffer, lest the input capacitance make the thing unstable.

Cheers, James

Use an inverting configuration so the 1G sees a virtual ground?

Cheers

--
Syd

Yes, that's what I was trying to suggest. That way the summing node doesn't have to move.

.-----[2M]--. | |\ | |\ | | \ | | \ >--[1G]--+--|- \ | | \ | >----+---|-1 >---> | / | / .--|+/ | / | |/ |/ ===

After a bit more sleep (;-), I don't think the op-amp's input capacitance is enough to destabilize it.

Cheers, James Arthur

That's about a wash, and I'd need to re-invert the TIA output with another amp.

The 1G resistor has some (currently unknown) shunt capacitance, and that makes a tau that any amplifier has to deal with. If the shunt C were zero, an ideal TIA would work.

If I have a reasonable capacitance across the 1G, then the lowside compensating cap (either amp config) gets up into the tens or hundreds of pF, where I can get a reasonable selection of compensation cap choices.

This is just a freebie current monitor so the customer can check that the pulse is working. It's not worth extreme measures.

There are PCB techniques for reducing the effective capacitance across resistors. I want to increase it!

TIAs like this can have noise peaking and oscillation from the tau formed by opamp Cin and Rf. My issue is that the parasitic capacitance across the 1G will cause an unknown output overshoot, which could be compensated by a cap across the 2M, which improves the TIA stability, too.

But if I have to put a cap across the 2M, I may as well use the single-opamp circuit.

I've spent a fair amount of sweat trying to reduce the effect of capacitance on large resistors. Fixing the frequency response isn't hard--you split it into multiple sections, with each set of interconnected pads sitting over a small pour on the adjacent level, and slots cut under the resistor bodies to reduce capacitance and get the flux out. Then you drive the pours with a lower impedance resistor string.

From my usual noise-obsessed POV that doesn't fix the problem, because the current noise of the first resistor couples through the pad capacitance to the low-Z string, dumping the noise into the summing junction. So it doesn't help the SNR except at low frequency.

It also isn't so easy to do at 1200V.

A gigohm and a picofarad makes a zero at 160 Hz. The circuit would have to be pretty slow for that not to dominate.

Good idea if the resistance tempco is tolerable. Good cermet pots come in at around +-100 ppm, iirc. John has been talking about 1 ppm/V high-ohm resistors for awhile, so the requirements might be tighter than that. Getting the untrimmed capacitances fairly close to right would allow the pot to be a small part of the total resistance, which would help a lot. Putting a slot in the PCB under the big resistor would make its capacitance more predictable, so the pot could be smaller.

I suggest a variable bootstrap. Drive the cold end of the bottom capacitor from a small-value pot hung off the output amp. That would make the capacitance adjustable, and the tempco of the pot would be ratioed out.

Cheers

Phil Hobbs

No, you want to control it, so that the shunt will match reliably. So larger than strays, and strays to a minimum.

RL

You probably want a guard at the divider node, because that 1G resistor ought only to be driving a resistive load. That means a guard/shield driven by the buffer amplifier, of course, and that's a bit of positive feedback. The alternative is to use inverting buffer, making the node pseudo-ground (and grounded shield is adequate).

Two guards are penciled in; the HV side of the 1G doesn't really require one, does it?

I'm guessing that the 2010 resistor will have enough capacitance that I don't really need more. The compensating cap will be 500x that value, so 0.05 pF across R1 needs about 25 pF below to compensate it. I can get a reasonable spread of caps around 25p, like 16, 18, 22, 27,

33...

Maybe I should use two caps, to allow fine tuning. I could use a trimmer cap, but that's one more thing the test folks would have to deal with.

A pot in a c-bootstrap is better than a variable cap, but still needs adjustment in production.

How about a capacitance that's tuned by two resistors? That will have better resolution than selecting from caps that we have in stock.

I still don't have any idea what the stray C might be. I suppose I could measure it. Grumble.

A guard will reduce my high-voltage clearance. Besides, I *want* some capacitance across the 1G resistor. I think.

The Welyn 2512 1G resistor (oops, not 2010) is rated for 3KV, which is pretty scary as regards surface creepage, even at 1200 volts.

The major imponderable will probably be the poorly-controlled epsilon of the board. If you make the air gaps big enough, you can probably control that well enough.

Cheers

Phil Hobbs