New HV and HF probe design

[Doug Coulter]

We got started talking over there but it fits better here if it should work - or for that matter fail spectacularly.

My basic first pass idea was to have a string of perhaps 10 meg resistors and somehow add a little capacity across each via the mechanical construction - perhaps round plates at each end of each resistor and all in a dielectric. For the simple 10x case with a 1 meg output and 25 pf (these might have to change to make the distributed caps reasonable), you'd have a 10 meg and 2.5pf element (R and C in parallel) in series with the 1 meg to the HV.

This should scale to any number of 10meg + 2.5 pf caps in series. It doesn't handle strays to the sides....I want to be able to put this inside an 8" or so ID shield along with other junk (the stepup xfrmr and so on) and get flat to say 20 mhz and down to DC (and I know a 10x is 9 megs and 1 meg, but here that sort of error we can just calibrate out).

[Doug Coulter]

BTW, here's a handy perl snippet that computes capacity between parallel round plates.
Perl comes on Linux and MacOS, for the computationally challenged, you can download a version for Windows from ActiveState or Strawberry and fiddle it into working.

Code: Select all

#!/usr/bin/perl
use Modern::Perl '2015';

my $dummy;
my $dia = .3;
my $sqin;
my $spacing = .4;
my $k= 1.0;
my $pf;

sub usage {
   
say << 'END_OF_STUFF';

Program will ask for diameter of cap plates in inches,
spacing in inches, and dielectric constant.
If you just hit enter it will use the last value, or
whatever I initialized above.
Control-c to quit.

END_OF_STUFF
}

usage;

while (1)
{
say "input diameter of plates, inches:";
$dummy = <>; # read input line
chomp $dummy; # remove the newline
no warnings "numeric"; # don't bitch if it's now empty
if ($dummy > 0) {$dia= $dummy;}

$sqin = (.5*$dia)**2 * 3.1415927;

say "input spacing in inches:";
$dummy = <>;
chomp $dummy;
if ($dummy > 0){$spacing = $dummy;}

say "input dielectric constant:";
$dummy = <>;
chomp $dummy;
if ($dummy > 0){$k = $dummy;}
use warnings;

say "sq in: $sqin, spacing: $spacing, dielectric constant: $k";

$pf = 0.2244 * $k * $sqin / $spacing;

say "Gives : $pf picofarads.\n\n;"
}

[Doug Coulter]

My current thinking is to use double sided PCB to mount the resistors and be the capacitors across them. Looks like I get 15 stages onto a 9" by .85" board dimension (fits in 1" thinwall quartz tubing) and would use 3 stages of that for a probe that would go to 50kv and beyond. Just spent the afternoon laying that out in Protel's old TraxEdit and EasyTrax software for DOS (emulated here on linux), which is my favorite for drawing-intense work where schematic capture and part number demands of newer but much less capable drawing programs need not apply. The new stuff might know which pin is !WE on some chip, but....making 2 pf caps out of PCB and making sure the right spots have the right spacing to kill leakage and corona?
This should work even in air, but would be nicer enclosed - assuming one stage is built and can be made flat - I'll machine some fancy ends for the tubing with O rings to hold oil or whatever in.

FWIW, anyone who wants copies of that super old CAD software, let me know.

[Doug Coulter]

I built one (15 R's and C's) stage for testing. I got a couple weird results, but at anything higher than a few hundred kHz it looks great - Looks like my PCB capacitors are a little large in value - much more fringing and other parastics than I expected, but...with a 200k load (instead of one meg) the load capacity to make it flat at HF is between that of a 1x and a 10x scope probe (and don't believe their specs on that - this ain't Kansas (Tektronix) anymore, Dorothy. We're in the 10pf+ range at 10x...

Anyway, the artwork is in the postscript files attached if anyone wants it - I just print these on transparencies with my laser (using the same one flipped over to be cheap) and expose pre-sensitized board and go from there.

I actually wound up writing a program to make the PCB file after getting a bit frustrated with the latest dosemu making the mouse pointer's DPIs not match....turns out the format is pretty easy to reverse engineer.

PROBE.zip

The PCB art and stuff

(6.97 KiB) Downloaded 406 times

15 sections of eventual 45 probe test sample

I pulled out some real vintage heavy hitter test gear to check this stuff:

GR FTW - but my hernia....

Messy bench. The probe section is to the right of the GR in quartz pipe.

The gain isn't perfectly flat...varies from a 760:1 to around 600:1 from 1 mhz to 10mhz (I can probably fix that). There is obviously some effect from nearby things with ground or other signals on them...I'm going to try to make this work inside some kind of shielding - all this is non trivial. But it looks pretty good so far.

scope input and output

At the lower F's with pulses (say .5uS), it shows a crazy big-fast overshoot, followed by a much smaller but DC-accurate pulse shape (and then undershoot on the back side of the pulse). I'll get a scope trace when I next do the required lifting (that oscillator must weigh 60 lbs), but it's weird in that at higher frequency sines - it's just "right" and real close to the calculated theoretical value. There's nothing in the setup with the kind of parastic L or C that would have a big effect down there at low F....a mystery at this point.

But, enough progress that I'm debating whether to change the PCB art a little before making 3 of these, or just making 2 more. These have 15 10 meg stages with very roughly 2pf in parallel with each, PCB spacings are ~ 100 mils so 1kv or so per stage should be no problem whatever, especially if I use glyptal or oil.

[Bob Reite]

Or pot it in epoxy. Oil always seems to leak all over everything.

[Doug Coulter]

I hate oil at least as much as the next guy...
I'm pretty good with lathe-made plugs and o-rings, though.
At these impedances...Not sure what would be best, but at prototype time, the ability to get in there and change things beats having to start over..and then there's the heat dissipation (depending, this can eat a few watts). Else I'd use paraffin...Oil would convect.

CliffS at Spellman once told me what they use but I couldn't find the exact thing at McMaster. It was some kind of 2-part silicone. Real high bulk resistance is a must here. I'm suspecting that "soak-in" in the FR4 is part of the errors I'm seeing, gheeze.

I've made 2 more PCBs, I'm going to stuff them and put them in a pipe. One issue with impedances this high is that "arm waving(1)" in the vicinity affects results...so I'm looking at what I can do to shield at least some of it. Up near the tip, there's basically no way - a 50kv differential and short distance don't mix well. Nor would capacitive coupling into middle stages from the original source.

I'm still getting somewhat weird responses down in the 100's khz - it seems fine but a little up and down (15% or so relative divide error) at 1 -20 mhz, though it's rock solid 1-10 mhz. (with just one 15 element stage - we'll see when there are 3 of those).

FWIW, HCl and peroxide work really well as etchant in "who cares" ratios. I used muriatic from an old jug and a splash of 12 volume H2O2 (dangerously strong without gloves) and wow is it fast. But stinky - HCl fumes.

(1) is only supposed to work on Vulture Capitalists, but Politicians try to use it on us all the time, too.

[Doug Coulter]

Here are all 3 built and being bench tested. I had some fairly bad fall off at HF (10-30 mhz) till I put them in the zig-zag config shown. That's going to be interesting to package and shield (at least the parts I _can_ shield). Picks up some 60 hz hum as is, to be expected when I "only" have 40v of test RF in, so I have to set the scope really sensitive. At least that's around 10x more signal than I'd get out of solid state gear.

Probe on bench. 45 x 10 meg + ~~ 2pf per resistor

Seems current scope probes have lots more capacity than the old Tek ones...but those old Tek ones cost around the same as a scope today in == value dollars, so not a big surprise. I had to go down to 200k load instead of a meg to get the RC ratio right. No big deal as I don't care that much what the ratio is as long as it's in a decently large range and I know the number.

[Bob Reite]

Putting it in the zig-zag is giving additional capacitive coupling about 1/3 and 2/3 across the stack. Probably only 0.5 pF worth, but at the Z you are dealing with now.... No wonder you have all the issues with hand waving on your results.

[Doug Coulter]

Yep...This is pretty complex to get any kind of flat, but I might wind up accepting 10% or so error if it's known and I can make it a little less hand-wavy. It will, after all, be mounted in a fixed spot with pretty constant field around it - or at least one that can be re-calibrated against. I have it pretty flat now except for a couple odd 3-6 db bulges and dips at frequencies that make no sense (like a bump near 3 mhz almost 6 dB - nothing in there would seem to have the right reactances for that).

It looked like this yesterday and I took a movie which I haven't yet edited. Has a frequency sweep in it. It's pretty impressive if you blink at the right moment and miss the error. I've since improved the mount and will be playing with gimmick capacity in both directions from various spots in the chain if for no other reason than to learn why a couple odd things are happening (could even be a test gear issue...). I now have a kludge board on the output side with variable R and C as well.

Pre stiff mounting - that's quartz tubing but the ends need dealt with for corona issues too...

[Doug Coulter]

Looks like this now:

Mostly more rugged for a wimpy definition of stout. Adjustable R and C load on board at bottom.

You can maybe see what normal looks like on the scope. I have this set up for 400:1 on the dot for DC and it stays inside 10% error over most of DC-15 MHz or so.

Above that, the phase reverses (What????, it's almost a purely capacitve volt divider, and FR4 is supposed to be really good, as is that NPO trimmer on the output).

Bad phase and divide ratio, note frequency here.

But what's baking my noodle is this one - and it's fairly narrowband around 3.37 something MHz. There's nothing deliberately nonlinear in sight, nothing I think would resonate at such a low F (it'd take long length and lots more C) and generally, what the heck?

Note harmonics and wrong divide ratio, only a couple hundred kHz wide with this visible at most here.

Here's an FFT if that helps anyone. Looks like 6th harmonic mostly. Other windows make it stand out more, but...this one is my fave.

FFT of nutty distortion

A bit disappointing. I can easily read through big slope-y errors, even ignore a skinny peak. But I want to see just such little wiggles when they are real.
Any thoughts?