A commodity-based HV regulator: FlexHV

[Elvee]

I have described a relatively simple and undemanding HV regulator here:
CCS + Zeners or resistors for 450V reference , but it requires a LT1013 as a gain engine.

Although it is a relatively cheap and common opamp; it doesn't qualify as a commodity: in some parts of the world, it could already be a kind of luxury.

Unfortunately, the circuit cannot tolerate a LM358 AND remain universal: the output stage of the 358 is too problematic for that.

Here is an adapted version, especially tailored for the 358.
It requires an additional buffer transistor and a different compensation scheme, and the static regulation performances are somewhat less shiny, because the gain and accuracy of the LM358 are not as good as the LT1013's, but it remains a capable regulator.




As in the previous circuit, the output voltage can be set with a single resistor.

The reference circuit also uses the AutoShunt principle, meaning many different types of references can be used: IC, LED, homebrew bandgap or even thermistors if you fancy it.
In my prototype, I have used a white LED, because it has ~the same voltage as the 2.5V shunt reference, and shows its operation in a visible way, which is useful for debugging work.

Although tailor-made for a 358, the circuit will also accept a LT1013, and many other dual opamps including GND in their CM range, but if you use a LT1013, it makes no sense to use this version, as it is more complicated and brings no benefit.

Here is the prototype in operation:





And here are the transient responses (with a static load of 12K and an output voltage of 215V).






The negative transient response is cleaner than the positive one, which is logical for a 1-quadrant regulator.

As its predecessor, it is unconditionally stable, and suffers the same drawbacks of minimum output voltage and 5V dropout.

For a system supply, this shouldn't be a problem, but if you want to build a lab supply, a scheme like Elektria is preferable.

Note that if you do not like ICs, I have also described simplistic, fully discrete regulators, but more than one resistor value needs to be changed if you want to significantly alter the output voltage:
Simple HV series regulators

[Ketje]

Couldn't it be done like this ?
No need for schottky and no opamp with input to ground.
Mona

[Elvee]

Yes, it is a functional possibility, but I didn't retain it despite the common-mode advantage: the main problem are the transients transmitted by C8/R18: clamping them to a relatively delicate reference IC is not a viable option, meaning a schottky to the local -6.25V would still be necessary, plus probably a clamp diode to the local 0V (output).
In principle, in a securely connected system, such transients should not be present, but as we know (you seem to have a good expertise in this field), these things happen with HV circuits, and it is preferable to eliminate all possible weaknesses before they have the opportunity to bite you.

Another (accessory) reason for rejecting it was that the initial reference, as a source for the autoshunt principle, could have any voltage and be much less stiff than an integrated reference (thermistor, etc).
If the autoshunt is designed for a final voltage of 6.25V, the circuit is normalized and the formula remains valid without needing to alter other resistor values.

Having a second opinion like yours is always welcome anyway, and DIYers could decide to opt for it if they think it is fit for their purpose.
Like all my projects, this one is completely open, and everyone is free to add his/her contributionS

[Ketje]

Aha, yes i see
How about a battery, NiCad or a lithium cel ?
Mona

[Ketje]

Well, perhaps a battery is not a very reliably reference.
And i goofed up with the voltage and consequent resistor in the schematic i posted .
The voltage drop on R13 isn't 3V3 but 6V6 (2mA !) and R15 has to be 150k.
And maybe the transient clamping problem can be solved with a capacitor over U2.
Mona

[sgrossklass]

What do you reckon, would some Class A biasing for U3 be of use? Q2 base current can't be that much, now can it? Added a 330k pull-down to GND, it didn't seem detrimental at least... ~1 mA seems like a more sensible output stage quiescent current than the 50 µA it's running stock.
Also, how about a cap across R7 to reduce reference noise, and a bigger cap in parallel to C8 for some more noise reduction? Both seem to have the desired effect. Probably of more use with a zener reference; an LED, by contrast, seems to appreciate an extra RC filter capacitor for its supply for reduced ripple (in return it sure looks much quieter in sim).

[Elvee]

Originally Posted by Ketje

Well, perhaps a battery is not a very reliably reference.

It could be quite reliable: an alkaline cell has a voltage of 1.59V, practically zero-tempco, zero-noise and is durable provided no current is drawn.

What stops me is the need to properly handle the off condition: how do you prevent the battery from discharging into various parasitic junctions when the circuit is not energized.
This issue can be addressed of course, as it was in CMOS memories, but it means additional complication.

And maybe the transient clamping problem can be solved with a capacitor over U2.

Yes, this is a viable and workable proposition, and in the original circuit C3 did the job of bypassing the transients away from the IC

Originally Posted by sgrossklass

What do you reckon, would some Class A biasing for U3 be of use?

Probably: the problem is caused by the class A to class B transition, so if U3 always remains in class A, it should be eliminated.
The small signal BW of the LT1013 is similar to the LM358, and the LT1013 works directly, without additional buffer.

Q2 base current can't be that much, now can it? Added a 330k pull-down to GND, it didn't seem detrimental at least... ~1 mA seems like a more sensible output stage quiescent current than the 50 µA it's running stock.

Q2's base current is tiny and isn't a problem.
To bias U3 in class A would require additional current, and since the regulator is fully floating and fed by the voltage-setting resistor, any additional consumption would need to pass through the Rset.
I opted for 2mA, and it already was a difficult tradeoff because the power at 500V output exceeds 1W.
Remember that the current through R4, R5 and possibly R1 also has to be sunk through Rset, and that the regulator is meant to be universal, even for insane input or output voltages within the 20V to ∞ input and 12.5V to ∞ output

Also, how about a cap across R7 to reduce reference noise,

Yes, I mentionned this possibility in the AutoShunt Thread, but the gain will be modest, a bit more than 6dB with a 2.5V reference.
If the reference has a lower voltage, the gain would be more significant.

and a bigger cap in parallel to C8 for some more noise reduction?

A larger C8 would benefit the VLF noise, not the 100Hz ripple nor the audio noise.
I opted for 470nF because it does not imply an E-cap, but if VLF noise or output impedance is particularly important, it could be increased (with a LM358, I estimate the 100Hz PSSR to ~120dB. with a LT1013 it is ~140dB)

Both seem to have the desired effect. Probably of more use with a zener reference; an LED, by contrast, seems to appreciate an extra RC filter capacitor for its supply for reduced ripple (in return it sure looks much quieter in sim).

Zeners, LEDs, or any forward-biased diode could benefit from an additional bypass cap. Integrated references like the LT1004 generally do not like a direct bypass for stability reasons, but with 100Ω series, any cap would be tolerable and the DC performances would remain unaffected, so there is room for improvement if needed.
This circuit is broadly the discrete equivalent of a 3-pin VR, and similar improvement tricks can be used, except you can also act on internal nodes, normally not accessible in an integrated regulator.
Contributions like yours are welcome since they show the way to turn this GP regulator into a HV super-reg

[carlmart]

Am I wrong or this is basically a Jung/Didden super regulator adapted for high voltages?

[jan.didden]

Originally Posted by carlmart

Am I wrong or this is basically a Jung/Didden super regulator adapted for high voltages?

No chance! With so many parts you can probably build three J/D super-regs ;-)
If you want a J/D for higher voltages, this would work.

Jan

[Ketje]

Just for fun, a simplex version, no opamps
Mona