431 based screen supply

[Hearinspace]

For experiments with the C3M I built a screen supply based on the schematic for a higher current shunt reg shown on the longer 431/1431 data sheets - that of a 431 controlling a transistor that works into a load resistor. After blowing up a couple of 431's, it was time to try using LTSpice (which showed the same voltages I was getting with real parts).
I eventually got the posted supply to give me a variable voltage between 60 and 180 Volts (by varying R3) , but it looks kind of hokey and if I put ac into the output of the reg where G2 will be (top of R2) it seems to get pretty well zero noise cancelling. Can someone help me improve this approach or is it from the start not worth considering?
Thanks !

[Rod Coleman]

Hi Ian! I have been halfway to nowhere with single-431 circuits. The problem returns to the one you've observed - high output impedance! You could use a source-follower on the output, but it still doesn't make a very good regulator....

....in fact, for VG2 regulators, a divider plus source-follower is usually plenty good enough, and generates negligible noise. Zener string + RC filter (don't forget the R!!) or cold gas tube, both with source followers, work well if your Vg2 needs to be better regulated.

[Hearinspace]

Alright I guess, but I can't let go of the idea without asking a question first. . . . Why?
With the PNP transistor there I would have thought the output impedance fairly low - at least before I added all those resistors so it would work in spice.
I've used this circuit with the BJT under the cathode of a triode and it worked well. Yet when I build it as pictured it simply doesn't work. Why?
My understanding is that it can put whatever current it needs to through the load resistor - netting the voltage required by the divider on the ref pin of the 431. But it didn't work. Why? (OK, that's three whys - your turn [please])

[Rod Coleman]

OK Ian, I will try!

the limitation is that with PNP shunt configuration, the base voltage cannot exceed the rated voltage durability (36V or part-dependent limit). Up to that value, the circuit will work as expected.


In your circuit, are you taking the PNP emitter as the output? you've drawn the base directly to the drive circuit, so that the 36V limit will still apply. Adding some resistors to pot-down the voltage safely is not satisfactory, because increased effective R will move the open-loop pole of the regulator down in frequency, which the control-loop will not be expecting - leading to oscillation. A real LT1431 has a COMP pin, so you may be able to add a cap there and compensate the regulator... but a small bandwidth regulator is not what is needed, I suspect.

If you really like the circuit, you can stack 35V regulators one atop the other to get the voltage you need, at the price of complexity, and zeners protecting each stage - and that is probably the only way to maintain the full bandwidth of the regulator.

With a single stage, you could try cascoding the output: add an NPN (MPSA42):
- start with the Figure 19 above:
- NPN base to +20V
- emitter to 431 output;
- collector to PNP's base/base resistor.

Might work quite well, because the combination-transistor on the output will have a fairly normal bandwidth. But, be prepared to compensate if necessary. The LT1431 can probably handle RC networks on its COMP pin, which will form a pole-zero: ideal for squeezing the best bandwidth out of the circuit.

If you use a different 431, please be wary of the models - dynamically, they're a bit off.

[aardvarkash10]

try this... or do a miada and be done with it.

[Rod Coleman]

Alright, I can't resist trying for myself!

The simulated performance is very decent - see the 30V (!) of ripple here is rejected to vanishing. The only oddity is that the current source at the input needs to be 200mA to make it work properly.

Might have to build one of these!

[Hearinspace]

Ack !! 200 milliamps? Are you serious?! And now we need a 20V supply as well?! No way! Next you'll have me running a gas generator to power it!

It's something like 10 or 12 years since John Broskie published his Hybrid circuit with the tube section buffering the high voltage, leaving the 431 to provide merely the cathode bias. It worked pretty well.
I was thinking of the BJT in Fig 19 as a substitute for the triode. I was wrong I guess, but surely there is a way to use a BJT/FET to replace the grounded grid triode in Broskie's circuit?

[Hearinspace]

Ack !! 200 milliamps? Are you serious?! And now we need a 20V supply as well?! No way! Next you'll have me running a gas generator to power it!

It's something like 10 or 12 years since John Broskie published his Hybrid circuit with the tube section buffering the high voltage, leaving the 431 to provide merely the cathode bias. It worked pretty well.
I was thinking of the BJT in Fig 19 as a substitute for the triode. I was wrong I guess, but surely there is a way to use a BJT/FET to replace the grounded grid triode in Broskie's circuit?

[Rod Coleman]

The 200mA was a problem for you to solve! it should not need it. I have found that LTspice behaves oddly with many common-base stages. And the 20V can come from a divider.

The Broskie circuit is almost the same approach. If you take the circuit I drew above, remove Q2 and its resistors, and connect the NPN collector to the output, it will work exactly the same as Broskie's.

Still, if you use it like that - whether with NPN or the triode - the performance will be quite poor, again because of fairly high output impedance (eg the Ra of the triode open-loop). The closed loop can give lower output impedance (at low frequency), but the open loop poles remain, and can affect stability if you need to add output capacitors.

If adding the PNP can be made to work, it will give low open-loop output impedance, which makes for a better HF performance.

The 20V supply can be 5V to 30V at 1mA, with RC filtering only.

[Rod Coleman]

BTW, if you really want an SS version of the Broskie, just slide a Supertex DN2540 in place of the triode.

Gate has 220R stopper, connect to GND. Source to 431-K. Drain to regulator output.

The depletion mode nature of this FET means it is the only transistor that can do this without the 20V cascode voltage.

EDIT: Cancel that. The range of Vgs (OFF) for the DN2540 is as low as -1.5V - which means you have no guarantee that the 431 has enough working voltage.

A normal enhancement mode FET with divider bias (or even battery bias) would work though.