Looking for advice to create/implement "grid driver" for vacuum tube RF PA

[wb0gaz]

I have a need to implement a gate driver-style circuit that will match 3V logic level (square wave) signal to the voltage and charge/capacitance of a vacuum tube (specifically, a triode or pentode) used in common cathode (i.e., grid driven) configuration as an RF power switch (it will end up being an RF generator, whose output is dictated by logic square wave from a simple synthesizer IC.)

This is an initial exploration (so I'm looking for general concepts and possible resources) as the tube and operating conditions have not been selected. Operation will be up to 50 MHz, input (grid to all other element) capacitance probably tens of pF range, and there may be some grid current drawn during positive-going part of the cycle (probably no more than tens of milliamps.) RF voltage as "seen" at the grid would be as much as a few hundred volts peak-to-peak.

The (likely) required RF voltage seems to make an IC gate driver impractical (typical FET gate driver is called on to drive at most tens of volts peak-to-peak, but with a much higher gate capacitance/charge and no expected gate current), but I may be searching improperly.

Grateful for any suggestions, ideas as to where to look for readings or circuit examples, and of course any requests for clarification of problem statement.

Thank you!

[Andy Chee]

Maybe you can adapt this circuit topology to suit your application?

https://sound-au.com/valves/design.html#s10

[Roehrenonkel]

Hi wb0gaz,
 
maybe a step-up transformer would work (in a limited frquency-range) for your application.
What tube (hundreds of volt grid-drive) =8-O have you in mind?
What anode-voltage? Kilovolts even?

Take care, good luck

[wb0gaz]

Thanks for the replies!

Andy - the linked article, while focusing I think on audio applications, is worth studying - PNP/NPN complimentary pair is what I've seen in some gate driver schematics, but I wasn't at all sure how to drive the pair when originating from the 3.3V synthesizer output. There are presumably "dead time" issues that need to be addressed, so that the two output devices don't try to be in conduction or transition at the same time, and the device selection would need to be revisited due to the ~1000-10000X higher operating frequency.

Roehrenonkel - you hit on a "unstated requirement" error in the original posting - I would want the grid driver to be functional across a fairly wide spectrum (at least 1.8 MHz to 50 MHz - single tone), so that would seem to rule out tuned circuit elements in the grid driver signal path. As for device(s) being driven - at the outset thinking 6146(s) or sweep tube(s) (single ended, parallel as needed, not push-pull), so 600-1100V on plate(s), with attendant fixed regulated, current limited screen voltage as needed.

In a nutshell, I'm interested in whether a single-ended FET "class E" type output stage - imagine a IRF510 switching FET being fed by a garden-variety gate driver at RF - could be implemented with vacuum tube(s), provided that a suitable gate driver (grid driver) could be realized.

Again, thanks for both replies - both very helpful and definitely press the idea forward!

Dave

[mag_therm]

Hi wb0gaz,
I have a homebrew Linear amp 45Watt for Ft8 , up to 14MHz
It uses high voltage SiC Mosfets with a drain voltage supply of 200V.
The drive boards are surface mount wideband amplifiers using "upside down " complementary BJT pairs like Fig9 in Andy's link
The RF circuit is more minimal than Fig9

I am presently building another set of boards slightly modified for 21 MHz  with intention of using sweep tubes in place of the Mosfets.
While hoping to keep the amplifier wideband (untuned), at >= 21 MHz neutralization of Ccb may be necessary on the driver output pair.

Looking at 6JS6 which are low cost compared the the similar, more popular 6JZ6, I see that Compactron tube bases may be coming hard to find.
I think with transistor drive it is a good idea to use tubes with top-cap anodes so the drive boards are physically spaced from the output.
6JS6 data sheets show a grid Eg of -30V for near cutoff, so peak grid drive would be about 28V pk to avoid grid current.

My circuit is push pull . One reason is the FCC out-of-band limit of -43dBc; P-P has inherent reduction of the 2nd harmonic.
I have adjustable bias for each side which allows trimming to minimize the 2nd Harmonic.
If the wideband amplifier inherently has high 2nd harmonics, the output filter becomes complicated and more lossy.
I presently have a surface mount double Tee constant K output filter which is not intended to attenuate the 2nd, it does mainly the 3rd and 5th.

Happy Toobing...
regards
KE8UZF

[wb0gaz]

Hello mag_therm,

Thank you for your time to post reply, your experience is very helpful to my learning!

I plan use 31LQ6 tube(s) at the outset because I purchased some long ago (new) at very low cost. They are somewhat larger plate dissipation than 6JS6, and are 9-pin compactron vs. the 12-pin compactron of the 6JS6. Both socket types (9- and 12-pin compactron) seem still available (try searching "dreaded online auction site"), the 12-pin seems more common, so you may find solution there.

Your point about bias cutoff voltage is a good reminder - thank you.

For my application, I would bias the tube to about 2x negative cutoff (per datasheet with experimental verification based on the operating screen/plate voltages), so with application of DC blocked RF from the gate/grid driver, the positive-going crest of the RF drive would be just at cutoff (hence tube remains cut off).

In parallel (via RF choke) I plan inject DC bias (positive) that would counteract the negative bias and as the DC bias increases, push the tube into conduction (out of cut-off) with positive crests of the RF drive. Idea is to control output power as a function of DC bias, with RF drive held constant (voltage). When resulting DC bias (fixed negative -2X cutoff and variable positive +1X cutoff) hits -1X cutoff, the tube would not be driven into grid current (so something like class AB1 except without idling current.)

Anyway, that's the idea at this point...

[Andy Chee]

Thanks for the replies!

Andy - the linked article, while focusing I think on audio applications, is worth studying - PNP/NPN complimentary pair is what I've seen in some gate driver schematics, but I wasn't at all sure how to drive the pair when originating from the 3.3V synthesizer output. There are presumably "dead time" issues that need to be addressed, so that the two output devices don't try to be in conduction or transition at the same time, and the device selection would need to be revisited due to the ~1000-10000X higher operating frequency.

Component changes will definitely be required, but the basic totem driver topology should be satisfactory to drive your valve.

Dead time is more of an issue with PWM power output stages, rather than driver stages.

Think of the driver as a miniature Class AB output stage. Yes, both transistors may conduct simultaneously, but the transistors won’t be completely turned hard on at that crossover point. If you want you can even re-bias the circuit for Class B and have zero crossover, depending on your required distortion specifications.
 

[T3sl4co1l]

You'll have a hard time finding BJTs anymore, that handle that kind of voltage and bandwidth.  Even salvaged CRT drivers (typically fT ~ 1GHz, Vceo ~ 100V) won't cut it -- you probably need 150-200V drive here.  2SC3503 is basically the only holdout, and while it is fast, it's not quite that fast.

Driving it wideband, also dictates quite a low impedance: a typical sweep tube might have 30pF input, so needs a grid-line impedance of 106 ohms or so.  You can get a little peaking from a resistive source into a capacitive load (30% or so), but not a crazy amount, even with a high order peaking network.

Example: I built this back in, uh, 2015ish was it?:



It's a proper 4-stage distributed amplifier.  I'd have to look at it to see what the grid line impedance was, but the terminator I think is 4 x 470Ω in parallel, so close to 120Ω.  I got about 30MHz out of it:



(input return loss measurement; never was able to get that final HF pole in line with the rest, which would've brought in a couple more MHz BTW, but adjusting these things is a PITA)

Bandwidth would've been higher if I went for a lower maximum power output; I used a singly-terminated plate line, which halves the bandwidth and doubles power (since it's not dumping backwards power into a termination resistor).  Pushing into the 60MHz range instead, it would've been even harder to tune, or maintain stability even, due to electrode inductances being a real problem up there.  And keep in mind, sweeps do not have particularly low reverse transfer capacitance; the screen grid, and electrode-connection area, make poor shields.  You'll be much better off using RF types like 6146 (preferably -B), but you'll pay for them, too.

Or radial-beam or even planar types, but besides being rare, they're harder to use (special sockets, forced air or more).

It's much less interesting, but far more practical, to simply go out and buy a MRF-whatever power transistor and be done with it; even plain silicon types have bandwidth beyond a GHz, and low enough Crss or s12 to make use of it too.  The circuit is simple, just a pair of bias tees, one rated for enough current to supply the drain, the other biased preferably with temperature compensation for stable biasing.  Impedances are low (few to tens Ω) so you do need to design the matching transformers a bit carefully, but that's standard TLT work.  Or use higher voltage types like MRF300 family for example, at the expense of lower bandwidth in part due to the cheap but poor choice of package, but capable of direct drive into 50Ω lines at reasonable power levels.

I also tried a pulsed test, in case sweeps or even those bizarro pulse regulator types perform any better at extreme cathode currents (which, in analogy with radar "hard modulator" types, should be permissible at modest expense to operating life):
https://www.eevblog.com/forum/projects/how-fast-are-tubes-for-switching/msg4676896/#msg4676896
as you can see, the impedances are still pretty high, grid current now extracts a serious toll on cathode current ("hFE" quickly drops from a cool ~infinity for Vgk < 0, to low 10s for modest forward bias, to single digits for strong forward bias), and if you're trying to get high power out of the thing, you need to boost Va a lot, further costing impedance and thus bandwidth.

SiC MOSFETs are probably the best current option for pulsed operation in this regime, i.e. 10s of kW and ~ns; you could make a quite reasonable EFT generator, for example, using a small handful of them.

Tim

[wb0gaz]

Tim - thank you **VERY MUCH** for the high level of detail. I will need to dedicate time this weekend to reading through; I may/probably have follow-up questions. I don't have any need to implement the driver with bipolars; FETs should be adequate, however, that will require some redesign I'm sure. The gate/grid voltage environment  looks like tens of volts, not hundreds (thinking from last comment the greatest negative excursion at the grid(s) would be -2x cutoff, maybe 100V or so, combining the standing bias of -1x cutoff and RF excursion of +/- 1x cutoff.) Designing a totem-pole driver operating in negative voltage regime will be new to me for sure!

[Andy Chee]

Designing a totem-pole driver operating in negative voltage regime will be new to me for sure!

That's where the transformer coupling comes in.  Your totem output (or single-ended or push-pull) can operate from whatever low voltage you have available, and drive the primary of the transformer to create sufficient peak-to-peak swing in the secondary.  The secondary of the transformer is connected to your valve's grid circuitry, and is biased to whatever voltage is necessary to cut off your valve.

Come to think of it, I've just thought of a junk box project along your lines (AD9850 DDS VFO driving a valve tube) so will be interested in what you eventually come up with.

[mag_therm]

Designing a totem-pole driver operating in negative voltage regime will be new to me for sure!

Come to think of it, I've just thought of a junk box project along your lines (AD9850 DDS VFO driving a valve tube) so will be interested in what you eventually come up with.

A bit off topic, but here is how I use a AD9850 DDS to drive Grid 2 of a pentagrid oscillator to injection lock an old HF receiver.
Untuned, it works from 1.8 to 30 MHz However to hold lock (for days or weeks) , the tube has regulated heater and plate, and a temperature controlled fan.
https://app.box.com/s/tcnq8ni9nne31wtzgjoge1s89fi46re6

[wb0gaz]

Thank you, Andy and mag_therm, for the added info!

On use of a transformer (but not necessarily because of using a transformer) - I now recall a tube operating in class C is biased beyond cutoff, then grid driven hard enough to overcome the prevailing bias. Is there any benefit in operating a tube in class E mode (like a FET would be in RF generator), or does that even make sense with a tube?

Also, going back to previous post - this generator project could of course be done with switching or RF power FETs (and from any practical perspective, that's the way to go), however, I'm focusing on tube design in this case because of an interest in whether tube(s) could be used for efficient RF generation (and in any event, FETs usually lack the nice, warm glow of tubes!)

[mag_therm]

For the grid characteristics of your 31LQ6 (which are not well shown in sweep tube datasheets) you could build a test Colpitts oscillator with a 31LQ6.
(A blocking capacitor is needed between anode and the Colpitts coil)
With adjustment of the feedback capacitor, the oscillator will start and run into "Grid Self Bias" where the grid swings up to a positive level, drawing grid current.
The grid then acts as a half wave diode rectifier and the grid resistor and Colpitts coil will charge down to a steady negative level.
With a 2 channel 'scope, the operating points, anode versus grid voltages can be obtained.
This oscillation mode is  stable if set correctly; as the anode AC voltage increases, the Colpitts coil will charge to a more negative steady state, so it is a negative feedback system.
While getting the feedback capacitor correct, it is best to put a current limiting resistor in anode ( or use a current limited supply)
That is in case the oscillation drops out and there is no negative grid bias.

[wb0gaz]

31LQ6 - can refer to 6LQ6 for specifications (differs only in filament voltage; these were intended to be part of series filament chain in a color TV receiver; I only specific that tube because I found a few very inexpensively long ago.)

Current limiting on plate ... yes good catch ... high-side current sensor in plate supply would be needed but never seen one (as an IC) with ~1kv withstand voltage. If I use resistive voltage dividers (2) to step down to a common IC sensor inputs (2), resistor precision and thermal stability probably renders impractical as any resistance difference in the two paths would create large difference at the current sensor inputs.

Will need investigate further plate current monitoring and fault prevention.

[mag_therm]

I  used 3AG 500mA fast glass fuse in the +200V drain (plate)  supply. But I should use a DC rated fast fuse like Bussman KLM-1/2 ( 600Vdc 500mA)
Available online they are a bit expensive and need a clip in holder.
I think not worthwhile to try to develop electronic protection.

The big 40 kW Triode Colpitts I worked with briefly (just before they went obsolete) had a current sensing relay on the 480V 3ph primary of the rectifier.
-Not very fast at all.

[T3sl4co1l]

On use of a transformer (but not necessarily because of using a transformer) - I now recall a tube operating in class C is biased beyond cutoff, then grid driven hard enough to overcome the prevailing bias. Is there any benefit in operating a tube in class E mode (like a FET would be in RF generator), or does that even make sense with a tube?

Well, you'll need a damper diode at least.  Note that class E produces a flattened waveform because the amplifier device is switching, and in particular, it switches on for negative current too -- think FET body diode.  You also need screen current limiting (preferably instantaneous, but that's nowhere near feasible at this frequency so average it must be), or use triodes only (at expense to saturation voltage, and, much more drive power required).  Otherwise, you need to operate in such regime where plate current is largely positive, which greatly limits load impedance range.

Sweep circuits themselves were class E, so it's not impossible.  That's just a very constrained application: no load variance to speak of, so it can be tuned just right.

Class E offers somewhat better efficiency, but you'll find heater power is always a significant fraction of the total and a more traditional class C amplifier (maybe driven into saturation/compression to some degree) will probably be more flexible, while being tolerant of load variation, and giving acceptable efficiency.

Obviously, these all need harmonic filtering at the output, and a tuned load is best.  A tuned load also provides impedance matching, so you can use a pi network for example, no transformer required.

Current limiting on plate ... yes good catch ... high-side current sensor in plate supply would be needed but never seen one (as an IC) with ~1kv withstand voltage. If I use resistive voltage dividers (2) to step down to a common IC sensor inputs (2), resistor precision and thermal stability probably renders impractical as any resistance difference in the two paths would create large difference at the current sensor inputs.

Easy, use one of the transconductance type ICs and cascode it:



Since there's a bias divider too, just chain them down to handle arbitrary voltage ratings.  P-ch aren't available in as diverse ratings as N, so it could take a few stages, but it's always doable.  Make sure Vgs is constrained on each one; probably the divider is backed with capacitors so AC (startup and short circuit conditions, ripple) divides evenly as well, and use G-S zeners to ensure drain capacitance doesn't mess up the next one down, etc.

Also, since your power supply is probably custom, might as well use a single-winding type and ground-return sense it.  This is trivial whether iron-core or SMPS.  (Not trivial if you're using a boost or something like that, but there's not much you'd want to boost from, let alone in a single stage, to that kind of voltage. And isolation is almost certainly required, so might as well make it a flyback or resonant supply.)

Tim

[wb0gaz]

Thanks, Tim --- now you've provided ample homework assignments so I'm going to cease asking questions for a little while until I catch up on the info you've kindly provided!!!

Dave