The high-voltage collection

[Elvee]

Here is a suite of circuits particularly (but not exclusively) useful to glass-lovers, mainly regulator circuits.

These are not top-class circuits, but they all share one common feature: an excellent effectiveness to investment ratio.

They are also mostly based on N devices, meaning an easier sourcing of parts, and a possibility to convert them partially or totally to vacuum.

An improved current source that can also be converted to 2-wires for floating operation:
Improved current source/sink

Another version, complementary but having a very wide voltage compliance:
Improved 2W current sources (II)

A robust and flexible series HV regulator:
Simple HV series regulators

A simple and crude shunt regulator, no more than a giant Vbe multiplier, but extraordinarily effective at hum elimination and wideband shunting:
No frills, cheap 'n crude HV shunt-reg/hum-killer

In fact, combined with just a passive resistor, it can still outperform a classic, state-of-the-art CCS + shunt reg.
The first pic shows the behavior when subjected to 90Vpp of ripple at 100Hz, with a 1.2K ballast resistor.
A typical classic regulator is shown as a reference.
The following pic is the same in the F-domain.

Note that the regulation performance is rather poor, and has the -0.3%/°C tempco of a typical Vbe mult, but the goal here is to clean up the supply, not to achieve an accurate output voltage.

Finally, an outline of a very high voltage driver, that could easily be converted to tubes, in order to drive directly large ESL or plasma speakers:
Self-cascoding, a tool for creating ultra high voltage building blocks:

[Sch3mat1c]

Why don't people just use a TL431 and forget about it? They're cheaper than LEDs, have more gain than several BJTs not even counting support circuitry (bias resistors, bypass or compensation capacitors if applicable), and are essentially a three-terminal op-amp with temperature compensated offset voltage. They're easily one of the most useful three-terminal devices out there.

I wrote this a while ago also:
New PNP HV tranny for anode CCS

Also commented on the "self cascoding". Despite its drawbacks, it should be useful for some limited applications, when you just have to drive kilovolts of something.

Tim

[Elvee]

Originally Posted by Sch3mat1c

Why don't people just use a TL431 and forget about it? They're cheaper than LEDs, have more gain than several BJTs not even counting support circuitry (bias resistors, bypass or compensation capacitors if applicable), and are essentially a three-terminal op-amp with temperature compensated offset voltage.

For voltages/currents/powers smaller than 40V/150mA/1W, they undoubtedly have a number of advantages, but as soon as you go beyond, you'll need discretes around anyway, and the configuration might not be ideally suited to all situations.
Some might require high voltage PNP's or P-channel, because the "sex" of the 431 is basically N (a NPN having a very sharp 2.5V Vbe), others may exceed its capabilities: the shunt regulator example regulates from 100µA, which is difficult to achieve for an isolated 431, even more when there is support circuitry around.

Anyway, I do not claim my proposals are a universal solution to any existing or conceivable problem, they are just one more piece in the "toolkit", I just hope they can be useful to some people.

If you have effective and original circuits based on the 431, why not share them?

[Sch3mat1c]

I was hoping the "voltage reference" motif would've been self-suggestive enough, but I suppose I can provide some examples anyway. The obvious solution is low side, which people use with LTPs and such. It should be obvious that a tempco as awful as a transistor's Vbe will directly affect the gain of such a stage, and especially in a tube amp, where heat is in no short supply, the amp will keep on drifting for the entire time it's heating up. A compensated reference is the obvious solution:



The high side (source rather than sink) is harder to do, and you are correct noting it's "NPN" and there is no "PNP" version. However, a workaround can be made, at some expense to saturation voltage, with the addition of one more transistor:



The additional compensation components Rz and Cz form a pole-zero compensator on the BJT which complements the dominant-pole compensation of the TLV431; together they should roll off smoothly from ~100s Hz up to the TLV431's f_T. As before, the values shown are TBD, and it may work without them; I would be surprised, however, if it works without any compensation at all ranges of current, bias and temperature. The compensation will impact output impedance slightly, but the MOSFET will do a fine job on its own, as has been observed before; and in the end, there isn't much we can do "up here" to impact its drain capacitance, which will inevitably dominate over 10-100kHz anyway.

Just because there's extra stuff in this one, and no reasonable intrinsic method for restraining gate voltage, I added a zener diode to protect the gate. In practice, 2N3904s avalanche in the 80-100V range, so I'm expecting 3906s do the same, which is no help here (gates typically rupture past 40-60V); the TLV431 is only rated for 30V, so the first circuit was passable without a zener (assuming Vgs(max) around 20V with a "you might be lucky" at 30Vgs). It could likewise be improved with a zener if you're particularly paranoid about gate protection.

All together, this circuit is certainly much denser than the previous, and also more than most any other circuit on the web. However, it will have guaranteed precision, as specified, within the range of bias current required by the TLV431 / 2N3906 pair (which should vary by +/-22uA over temperature). The same cannot be said of any other circuit I've seen, which have substantially worse tolerances on output current (due to Vbe or Vgs(th) variation with temperature).

Tim

[Elvee]

Originally Posted by Sch3mat1c

I was hoping the "voltage reference" motif would've been self-suggestive enough, but I suppose I can provide some examples anyway. The obvious solution is low side, which people use with LTPs and such. It should be obvious that a tempco as awful as a transistor's Vbe will directly affect the gain of such a stage, and especially in a tube amp, where heat is in no short supply, the amp will keep on drifting for the entire time it's heating up. A compensated reference is the obvious solution:

This circuit could include the same trick as my improved CCS V1 to make it 2-wire.
This would allow for a universal, unisex circuit, with a N-channel output and without the voltage translating complication of the second circuit.

A suitable resistor simply has to be added between the 431 Adj. and the + side of the CCS.

[Sch3mat1c]

That won't work while preserving the high impedance. Perhaps you could add series resistance below the MOSFET, raising the source voltage enough to power the TL431, much as I did for the PMOS. Unfortunately, you still need something to drive the MOSFET gate, although a depletion mode device would work.

Tim

[Elvee]

Originally Posted by Sch3mat1c

That won't work while preserving the high impedance.

Why not?
Obviously, the exact zeroing of the conductance will depend on accurate matching of the resistances ratios, but with good matching, the performances should be excellent.
If I can find an acceptable model for the TLV431, I'll try to sim it.

[Elvee]

It works, except for the compensation that made it oscillate like mad, whatever the values (including without the correction trick).

I had to change the strategy. Probably not optimal either, but sufficient to test the circuit. Model is from TI, so it shouldn't be too fanciful.

Without too much tweaking, the DC resistance is ~40meg (for 10mA current):

[Sch3mat1c]

Capacitive current at that, so it works out well enough.

You want a cap or R+C from TL431 "anode" to "ref" (i.e., output to -input as an op-amp) for compensation.

Tim