Win's next 10kV project, a 1us ramp

[Fred Bloggs]

Joseph2k wrote...


Yes, harder... I can easily handle a design for 50:1 spread with
low-droop requirements, for an "ordinary" transformer, but for one
insulated for 10kV, I'm not so confidant.




Maybe, maybe not. What you get that's linear on a typical HOT is
the charging-current ramp, and not necessarily the transformer's
output-voltage, which doesn't have an ramp as a goal. Still, a
discharging flyback transformer could exhibit a constant output
current, making a linear ramp into a capacitor... So, I'm still
thinking about it.

Well what can you do in a transformer for a 1:12 1kV primary 10kVA rated
secondary in the way of leakage, magnetizing inductances and secondary
shunt capacitance? I still say it is a regulator function, not a servo
function.

 

[Tim Williams]

Maybe, maybe not. What you get that's linear on a typical HOT is
the charging-current ramp, and not necessarily the transformer's
output-voltage, which doesn't have an ramp as a goal. Still, a
discharging flyback transformer could exhibit a constant output
current, making a linear ramp into a capacitor... So, I'm still
thinking about it.

Might work if you clamp it to 10kV (er, 13kV?) and use a high inductance
charged to a chunky current, with F-all parasitic capacitance (that won't be
hard, at least, with all the insulation required). Then add more C, and
maybe some R and L, to tweak the slope and such. Point being, the
*unclamped* flyback pulse would be a decaying sinewave of a few hundred kV
or so peak, so it's just faked by using some approximated slope. Could
still be good enough though. You'll need more L for the slower slopes,
since dI/dt = V/L means current drops off noticably for longer periods.

Tim

 

[Winfield Hill]

Tim Williams wrote...

Might work if you clamp it to 10kV (er, 13kV?) and use a high inductance
charged to a chunky current, with F-all parasitic capacitance (that won't
be hard, at least, with all the insulation required). Then add more C,
and maybe some R and L, to tweak the slope and such. Point being, the
*unclamped* flyback pulse would be a decaying sinewave of a few hundred kV
or so peak, so it's just faked by using some approximated slope. Could
still be good enough though. You'll need more L for the slower slopes,
since dI/dt = V/L means current drops off noticably for longer periods.

Now you're *really* winging it. Crank the numbers.

 

[Tim Williams]

Now you're *really* winging it. Crank the numbers.

Well, that doesn't suprise me, but drawing it out, figuring 500pF gives 5A
in the inductor, final capacitance energy will be 1/2 * 10kV^2 * 500pF =
50mJ so for 10% SWAG, the inductor needs to store around a half joule,
rather chunky but not impossible. 0.5J at 5A is 40mH, pretty chunky for the
required parasitic capacitance I suppose, but tigher requirements have been
met in components..(?) Of course, most of half a joule has to be removed
from the inductor, but a fast diode can dump that into a capacitor for
recycling (or wasting, who cares).

40mH with 500pF is 7us quarter-wave, so 1us is certainly on the rising edge
part. If the slope isn't smooth enough, you can always increase L, to swamp
the dI/dt droop as I mentioned.

To use normal parts, you could wind a transformer, needing messy low leakage
inductance of course, and use a MOSFET or IGBT to switch it (1kV at 50A is
no problem, I've got a fifth of that (i.e., 200V) four feet away from me)
and a diode to clamp the overshoot. Otherwise, 5A by 15kV isn't impossible
for a suitable pentode.

I don't see anything patently wrong with it electrically, but the
particulars are pretty strict, capacitance, leakage inductance (if using a
transformer) and whatnot.

Or I could always ask you if you have any better ideas. But then, I could
also ask why you posted this thread. But I won't. ;-)

Tim

 

[Terry Given]

Winfield Hill wrote:


I looked through the catalog but those things do not behave like what i read
about. One specific advantage was faster switching that the spark gap
things.

static induction thyristors?

Cheers
Terry

 

[Tony Williams]

Well, that doesn't suprise me, but drawing it out, figuring 500pF
gives 5A in the inductor, final capacitance energy will be 1/2 *
10kV^2 * 500pF = 50mJ so for 10% SWAG, the inductor needs to
store around a half joule, rather chunky but not impossible.
0.5J at 5A is 40mH, pretty chunky for the required parasitic
capacitance I suppose, but tigher requirements have been met in
components..(?) Of course, most of half a joule has to be
removed from the inductor, but a fast diode can dump that into a
capacitor for recycling (or wasting, who cares).

40mH with 500pF is 7us quarter-wave, so 1us is certainly on the
rising edge part. If the slope isn't smooth enough, you can
always increase L, to swamp the dI/dt droop as I mentioned.

Yes, I did similar sums last week, from a slightly different
starting point, but producing similar numbers. The idea is
just about ok for a 1uS ramp, but horrendous for 50uS.

My starting point was that a sinewave is a reasonable
straight line from zero to about one third of Vpk. So you
aim as if generating a 4.5uS quarter sine with 21KVpk, and
clamp at 7KV, (with 35-off 200V TVS devices).

4.5uS quarter sine is a resonant frequency of 55KHZ. Taking
your 500pF, that would need a secondary inductance of 18mH.

21KV across 500pF is 110mJ. That would require the 18mH
secondary to have an Ipk of about 3.7A.

If the driving FETs were 1KV types clamped to 800V that
would be a 1:8.75 turns ratio. Primary inductance would be
235uH, charged up to an Ipk of 32A.

Now do the same sums for a 50uS ramp.

 

[Tim Williams]

The idea is
just about ok for a 1uS ramp, but horrendous for 50uS.

100H? Nahhh ;-)

So, you'd want to increase C. Downside: more stored energy means way more
inductor energy. So lesse, C = 5nF and L = 500H at 100mA... eh, okay,
nevermind...

Is there a reasonable way you could switch to a follower for the slower
sweep rates? Current demand is certainly lighter.

Tim

 

[Winfield Hill]

Tony Williams wrote...

If the driving FETs were 1KV types clamped to 800V that
would be a 1:8.75 turns ratio. Primary inductance would
be 235uH, charged up to an Ipk of 32A.

Now do the same sums for a 50uS ramp.

Well, I wouldn't say 32A at 800V is so great either...

I just won this little 5D21, a perky 20kV 15A baby for $6.50,
http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=6596274062
http://www.mif.pg.gda.pl/homepages/frank/sheets/061/p/P535-1E.pdf

Having paid for his old tube, maybe I can go on and purchase
the seller's "Genuine Broken Horn Rodeo Belt Buckle."

 

[Winfield Hill]

Tim Williams wrote...

Tony Williams wrote in message ...

The idea is just about ok for a 1uS ramp, but horrendous for 50uS.

100H? Nahhh ;-) [ snip ]
Is there a reasonable way you could switch to a follower for the
slower sweep rates? Current demand is certainly lighter.

I suppose, but it's better to simply reject the idea and move on,
don't you think? I've sketched out five ways to do it, and have
rejected three. The most appealing way is to use one 20kV tube.
I won an old 5D21 on eBay (there are lots of them out there) for
$6.50, and found a fellow who'll sell me a brand-new Eimac 4PR60C
for $175. http://www.eimac.com/catalog/170300.htm

One fellow reports on the net of getting these beasts to generate
a 10kV 10A pulse with a 10ns risetime -- pretty impressive, huh?

 

[John Larkin]

Tim Williams wrote...

Tony Williams wrote in message ...

The idea is just about ok for a 1uS ramp, but horrendous for 50uS.

100H? Nahhh ;-) [ snip ]
Is there a reasonable way you could switch to a follower for the
slower sweep rates? Current demand is certainly lighter.

I suppose, but it's better to simply reject the idea and move on,
don't you think? I've sketched out five ways to do it, and have
rejected three. The most appealing way is to use one 20kV tube.
I won an old 5D21 on eBay (there are lots of them out there) for
$6.50, and found a fellow who'll sell me a brand-new Eimac 4PR60C
for $175. http://www.eimac.com/catalog/170300.htm

One fellow reports on the net of getting these beasts to generate
a 10kV 10A pulse with a 10ns risetime -- pretty impressive, huh?

The implied capacitance is what's impressive!

John

 

[Winfield Hill]

John Larkin wrote...

Tim Williams wrote...

Tony Williams wrote in message ...
The idea is just about ok for a 1uS ramp, but horrendous for 50uS.

100H? Nahhh ;-) [ snip ]
Is there a reasonable way you could switch to a follower for the
slower sweep rates? Current demand is certainly lighter.

I suppose, but it's better to simply reject the idea and move on,
don't you think? I've sketched out five ways to do it, and have
rejected three. The most appealing way is to use one 20kV tube.
I won an old 5D21 on eBay (there are lots of them out there) for
$6.50, and found a fellow who'll sell me a brand-new Eimac 4PR60C
for $175. http://www.eimac.com/catalog/170300.htm

One fellow reports on the net of getting these beasts to generate
a 10kV 10A pulse with a 10ns risetime -- pretty impressive, huh?

The implied capacitance is what's impressive!

For a big 20kV tube that can do 20A, or more, it is impressive,
only 8pF output and 37pF input capacitance. Knocks my socks off.

 

[Winfield Hill]

Winfield Hill wrote...

Ancient_Hacker wrote...

I haven't found any of them, but 5D21 tubes are available from
WW-II days. Thanks for the suggestion, they look promising.

We need a -3kV to -13kV voltage ramp, and a good way to get that
is a constant sink current into the node capacitance. For example,
if the capacitance is 100pF, we need -1A for 1us.

The schematic of a high-voltage vacuum-tube solution indeed looks
simple, until I add the issue of dealing with the changing grid
voltage required to maintain a constant plate current, as the
plate voltage drops from say 13kV to 3kV. :-(

So, given my familiarity with MOSFETs, I keep thinking about them.

A full 10kV ramp would mean that cascode FET stages are required.
Here's an appealing simple two-stage cascode approach, capable of
creating a -3kV ramp using three 2sk1412 1.5kV 150mA MOSFETs.

.. -3kV
.. |
.. 1mA
.. -4.5kV ______ | out
.. | | +----------------------,
.. _|_ | | ___ |
.. --- 1mA | \ _|_
.. 2 Co | | | \___-3kV --- Co
.. |____|_______||-' ramp | 75-100pF
.. | | ||-, Q1 |
.. _ | | 10V | gnd
.. __| |__ | '--|<|--+
.. ---+---------- |---, |
.. | _ Q3 | | _ |
.. '--| \___||-' '--| \___||-' Q2
.. ,--|_/ ||-, ,--|_/ ||-,
.. | | | |
.. '-----------+ '-----------+
.. | |
.. Rs Rs
.. | |
.. '---------------+---- -6kV

The 1.5 and 3kV-rated 1mA pullup current sources are series
stacks of my favorite Supertex LND150 depletion-mode MOSFETs
that return their capacitive nodes to their quiescent states.

This is a fairly-simple two-stage circuit. But my efforts to
extend it to more stages have resulted in complicated messes.

 

[Jim Thompson]

Winfield Hill wrote...



The schematic of a high-voltage vacuum-tube solution indeed looks
simple, until I add the issue of dealing with the changing grid
voltage required to maintain a constant plate current, as the
plate voltage drops from say 13kV to 3kV. :-(

So, given my familiarity with MOSFETs, I keep thinking about them.

A full 10kV ramp would mean that cascode FET stages are required.
Here's an appealing simple two-stage cascode approach, capable of
creating a -3kV ramp using three 2sk1412 1.5kV 150mA MOSFETs.

. -3kV
. |
. 1mA
. -4.5kV ______ | out
. | | +----------------------,
. _|_ | | ___ |
. --- 1mA | \ _|_
. 2 Co | | | \___-3kV --- Co
. |____|_______||-' ramp | 75-100pF
. | | ||-, Q1 |
. _ | | 10V | gnd
. __| |__ | '--|<|--+
. ---+---------- |---, |
. | _ Q3 | | _ |
. '--| \___||-' '--| \___||-' Q2
. ,--|_/ ||-, ,--|_/ ||-,
. | | | |
. '-----------+ '-----------+
. | |
. Rs Rs
. | |
. '---------------+---- -6kV

The 1.5 and 3kV-rated 1mA pullup current sources are series
stacks of my favorite Supertex LND150 depletion-mode MOSFETs
that return their capacitive nodes to their quiescent states.

This is a fairly-simple two-stage circuit. But my efforts to
extend it to more stages have resulted in complicated messes.

Maybe use a toob for the cascode device (FET under cathode).

I've done lower voltage situations with a bipolar/op-amp under the
cathode of a toob.

...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona Voice480)460-2350 | |
| E-mail Address at Website Fax480)460-2142 | Brass Rat |
| http://www.analog-innovations.com | 1962 |

It's what you learn after you know it all that counts.

 

[John Larkin]

Maybe use a toob for the cascode device (FET under cathode).

Dang, I was going to say that, but I was too busy eating.

I've done lower voltage situations with a bipolar/op-amp under the
cathode of a toob.

Also fun: use a high-voltage rectifier tube, like a 1B3, and control
the filament voltage to modulate its conductance. A bit slow for Win's
app, I admit.

John

 

[Joseph2k]

Tim Williams wrote...

Now you're *really* winging it. Crank the numbers.

Winfield, i find a continuity error between my post and yours. I cannot
find the quoted test in my post. Will you help me understand this?

 

[Winfield Hill]

John Larkin wrote...

Dang, I was going to say that, but I was too busy eating.
I've done lower voltage situations with a bipolar/op-amp under
the cathode of a toob.

Yes, of course, that's the way to do the tube solution. I realized
that just after hitting SEND. I've often seen this done before, but
my memory was slow to reveal the answer.

However, the tube approach is still rather complicated, because the
"grounded grid" tube with its programmable current-sinking MOSFET
cathode driver (500V, 1A - not too bad) and its 50W filament power,
will all be sitting at -16kV. Hmm, that better all be inside a big
Faraday cage with rounded corners. Can't have corona discharge off
of our sharp-edged components! Maybe we can use a clear housing, or
add some windows.

A 5D21 or 4PR60C tube needs to have 50W of filament heating removed
(Eimac says by radiation and forced air, but there's no air chimney).
Damm, I hate to think of all the dust that'll collect on the -16kV
surfaces; maybe we can seal it, and use recirculating air with some
type of heat-exchanger. We can monitor the temperature. Hmm.

We'll need a two-way optical-fiber communication link (for DAC and
logic signals up, ADC diagnostic info back).

We'll need to make or buy a 75W, 20kV-insulation-rated power source
for the cage, mostly for the filament. We can use a dc-dc converter
inside the cage to get say 600V to 800V for the screen grid (the
cathode will be 100 to 350V above the "grounded" grid). Hmm.

We might need a pullup resistor or current source for the cathode,
up to the 600V supply, to allow for rapidly turning off the tube by
charging the MOSFET's drain. Hmm, ahem. No, that probably won't
be necessary, the tube's high current when running should do that
fast enough, the cathode capacitance will be under 100pF. Hmm.
That's only 200V/us at 20mA, maybe we'd better use one after all.
The operating range of the pulser will be about 0.1mA to 1A.

Well, it's still less of a mess than the cascoded-MOSFET solutions
I've sketched out so far... Hmm.

 

[Ancient_Hacker]

The schematic of a high-voltage vacuum-tube solution indeed looks
simple, until I add the issue of dealing with the changing grid
voltage required to maintain a constant plate current, as the
plate voltage drops from say 13kV to 3kV. :-(


Did you look at the curves for the 715-C at: http://tinyurl.com/akggd
?

The curve looks quite pentode-like-- quite flat. Might only need a
middling cathode resistor to flatten it out enough?

BTW these tubes are often on eBay for under $10. Of course, untested
at 15kV!



Regards,

grg

 

[Winfield Hill]

Winfield Hill wrote...

...given my familiarity with MOSFETs, I keep thinking about them.

A full 10kV ramp would mean that cascode FET stages are required.
Here's an appealing simple two-stage cascode approach, capable of
creating a -3kV ramp using three 2sk1412 1.5kV 150mA MOSFETs.

. -3kV
. |
. 1mA
. -4.5kV ______ | out
. | | +----------------------,
. _|_ | | ___ |
. --- 1mA | \ _|_
. 2 Co | | | \___-3kV --- Co
. |____|_______||-' ramp | 75-100pF
. | | ||-, Q1 |
. _ | | 10V | gnd
. __| |__ | '--|<|--+
. ---+---------- |---, |
. | _ Q3 | | _ |
. '--| \___||-' '--| \___||-' Q2
. ,--|_/ ||-, ,--|_/ ||-,
. | | | |
. '-----------+ '-----------+
. | |
. Rs Rs
. | |
. '---------------+---- -6kV

The 1.5 and 3kV-rated 1mA pullup current sources are series
stacks of my favorite Supertex LND150 depletion-mode MOSFETs
that return their capacitive nodes to their quiescent states.

This is a fairly-simple two-stage circuit. But my efforts to
extend it to more stages have resulted in complicated messes.

Here's one approach to extending the circuit. You can see the
complexity grows rapidly as cascode stages are added, because
a new current sink is required for each gate-biasing capacitor,
and these also have to be cascoded.

.. -3kV
.. |
.. 1mA
.. | out
.. -4.5kV ______________ +--------------------,
.. | | | |
.. _|_ | | ___ -3.0 |
.. --- 0.5mA | \ |
.. Co | | | \ _|_
.. -6.0kV ______ |____|_______||-' \___-7.5kV --- Co
.. | | | | ||-, |
.. _|_ | | | 10V | -4.5kV ramp |
.. --- 0.3mA | '--|<|--+ gnd
.. Co | | | |
.. |____|_______||-' __________||-'
.. | | ||-, | ||-,
.. __ | | 10V | | 10V |
.. __| |__ | '--|<|--+ '--|<|--+
.. --+---------- |---+---------- | --, |
.. | _ | | _ | | _ |
.. '--| \___||-' '--| \___||-' '--| \___||-'
.. ,--|_/ ||-, ,--|_/ ||-, ,--|_/ ||-,
.. | | | | | |
.. '-----------+ '-----------+ '-----------+
.. | | |
.. 3 R 2 R 2 R
.. | | |
.. --------------+---------------+---------------+--- -7.5kV

At this point, with only a 4.5kV capability, the MOSFET circuit
may still be more simple than the tube circuit, but extending
the scheme to 6kV would probably tip the balance.

If I could figure out a way to stack the gate-bias capacitors...

 

[Winfield Hill]

Ancient_Hacker wrote...

Did you look at the curves for the 715-C at: http://tinyurl.com/akggd
?

The curve looks quite pentode-like-- quite flat. Might only need
a middling cathode resistor to flatten it out enough?

Yeah, they're pretty flat, but we need it flatter, plus we need
it programmable. That's not so hard, however, with a 500V power
MOSFET in the cathode.

BTW these tubes are often on eBay for under $10. Of course,
untested at 15kV!

Right. I bought an old 5D21 on eBay for $6.50 this weekend, but
a new production Eimac 4PR60C made 8 years ago set me back $175.
Let me say I really appreciate your first suggesting the 715 --
that got me going in the right direction, and was a big help!

 

[Winfield Hill]

John Larkin wrote...

[ snip ] The most appealing way is to use one 20kV tube.
I won an old 5D21 on eBay (there are lots of them out there) for
$6.50, and found a fellow who'll sell me a brand-new Eimac 4PR60C
for $175. http://www.eimac.com/catalog/170300.htm

One fellow reports on the net of getting these beasts to generate
a 10kV 10A pulse with a 10ns risetime -- pretty impressive, huh?

The implied capacitance is what's impressive!

One can think of our experiment as simply requiring a 10kV 1us
ramp, sort of like this,
.. _____
.. \
.. \___

But it may be more instructive to look at it like this,

.. ___________________
.. ________/ \______
..
.. --------,,
.. \
.. \
.. \
.. \
.. \
.. \ -10kV/us
.. \
.. \
.. \
.. \
.. \
.. \
.. \
.. \
.. \
.. \
.. \
.. \
.. '-__________+10V/us

Clearly to get an accurate, useful 1us ramp, I've got to start
the charging current in say 25ns or less, and get it stabilized
in a similar time. So I find the 10ns story encouraging.