• WARNING: Tube/Valve amplifiers use potentially LETHAL HIGH VOLTAGES.
    Building, troubleshooting and testing of these amplifiers should only be
    performed by someone who is thoroughly familiar with
    the safety precautions around high voltages.

Jan's HV regulator


More to the point, I see a few potential problems:

From an AC perspective, the opamp operates in unity-gain, with 100% feedback, and to ensure stability the feedback path must introduce a minimal phase-shift between the opamp output and its - input, and this has to be true between 0 and ~80MHz, since no local compensation is present.

The gate stopper and the MOS capacitances create a first pole, meaning any other pole along the path will lead to instability.
Such a pole could be created by the protection resistors and a capacitive load for instance.
A quick, simplified sim shows that the regulator is marginally stable on a resistive load and becomes unstable with a capacitor of a few nF.
The reality could be somewhat different, but very much will depend on the exact load, the way it is wired, etc, and this is not controllable on the board itself, meaning it should be reasonably immune to impedance effects.
The + input connection also contributes to complicate the problem.
For these reasons, I think that a minimum of local compensation should be included.

Other problems could be caused by C4: it will be charged to the output voltage, and in case of a short, it will discharge through D3.
The energy will be low, but after some events a small diode like the 1N4148 might well fail.
Note that the "short" does not need to be a hard, physical one: the connection to a completely discharged bypass cap will have the same effect.
Less obvious is the case of load-dumping: there is apparently no good reason for such an event to occur, but in practice HV circuits can sometimes see this kind of situation, especially in an experimental/DIY setup.
If the output voltage increases brutally over the normal, set voltage (a cap discharge, short to another supply...) the whole regulator including the opamp will rise higher than C4, leaving the + input protected only by the 100 ohm series resistor.

For these reasons, it would be preferable to place two anti parallel diodes directly across the inputs, possibly schottky types.

In case of an output short, another component will be stressed: R10.
It will have to withstand a high surge current, and should be a pulse-resilient type.

Something optional, but cheap and contributing to the overall reliability would be the inclusion of a series resistor (preferably fusible-type) in the drain of Q8: up to 1K will have no impact on the performance and could save the day in case the MOS encounters the occasional rogue pulse
 
I am convinced that the regulator in itself is stable and doesn't need additional bypass, but it goes further than that: it might not tolerate a bypass cap directly connected across the output terminals.

The issue is common to high gain, high feedback amplifiers: for example, if you connect a 100nF capacitor directly to the output of an audio amplifier, upstream of the inductive zobel it will most of the times cause oscillations.

Voltage regulators are a particular class of amplifiers having a single-quadrant output, but they are subjected to the same general rules.
They are normally compensated in a way that makes them tolerant to capacitive loads, because that is the way they are normally used. Even then, a minimal esr value is sometimes required to ensure stability.
R6 is a very explicit series resistor, and it adds a zero in the response, but an additional, pure cap might cancel the stabilizing effect of this zero.

I have no idea about the way the circuit is going to behave in reality, because so many factors play a role, but having such a large R10 is like a door open to the outside world: internal loop stability issues will be influenced by the nature of the external load.
With a "hard" decoupling capacitor, the door is completely closed for high frequencies, and the regulator becomes just a black box delivering voltage and current, without further interaction.

Testing the response with current steps, together with various reactive loads could provide valuable information about the loop stability
 
Improving the stability can be as simple as opting for a more placid, some would say sensible opamp.


Here is the simplified loop gain sim (the exact details will certainly differ in reality, but it gives a broad idea of what to expect).
With a 47nF cap, the phase margin is -23°, meaning it will be unstable:

attachment.php


If the opamp is changed for a LT1677 (GBW=7.2MHz), the circuit becomes marginally stable, with a 8° margin.
Note that the performances have not been degraded, in fact the opposite is true: the 100Hz loop gain (dictating the ripple rejection) has improved by ~15dB:

attachment.php


With a small cap across the gate-stopper, this margin is almost doubled:

attachment.php
 

Attachments

  • Jan1.png
    Jan1.png
    109.1 KB · Views: 621
  • Jan2.png
    Jan2.png
    95.3 KB · Views: 647
  • Jan3.png
    Jan3.png
    95.7 KB · Views: 635
I think you are right: I probed it that way because the regulator is referenced to the output, meaning the gain of the MOS operating in common-source has to be taken into account, except that here the configuration is hybrid.

Although the regulator circuit is referenced to the output, the reference voltage is in fact tied to the ground, meaning it acts as a follower.

To take the whole situation into account, including common mode effects of the opamp, it is probably necessary to probe it that way:

attachment.php


Ideally, a more sophisticated probe would be necessary to include the loading effect on the output, but here with the simplifications it is probably sufficient, and including the opamp supply in the probe changes practically nothing, thanks probably to its good common-mode rejection.

The LF gain is substantially reduced, but the HF behavior remains essentially similar, and the negative phase margin is still present, it is even a bit larger.

The same remedies apply, and have a similar effect, and if they are combined to your fixes, it is probably possible to arrive at a satisfactory solution, but it has to be tested in practice, because tailoring the compensation around a 47nF load will certainly be non-ideal for all other values (it is an example I took randomly).

attachment.php
 

Attachments

  • Jan4.png
    Jan4.png
    105.6 KB · Views: 556
  • Jan5.png
    Jan5.png
    93.1 KB · Views: 521
AX tech editor
Joined 2002
Paid Member
In any case lowering R3 and increasing C1 improves phase and gain margin.

The value of 1uF for C1 was a practical value as to size/voltage limit (630V) on the PCB. Design target was 600V input and thus possibly 600V output.
With that part in mind, R3 was selected for best stability in several prototypes.

I did look at larger caps at 630V, but to my surprise 600V+ electrolytics are very rare.

Jan
 
The regulator has been designed to be stable without external output capacitance.
That is clearly the case
Adding output capacitance makes it more stable.
Facts cannot be refuted, thus reality has to be the ultimate test: there are many effects and parasitics involved, and a simplified sim cannot replicate the fine details.

The topology of a buffer based on a fast, high gain uncompensated amplifier rang an alarm bell for me, because when combined with a capacitive load, it is generally a recipe for instabilities, and the sim seemed to confirm this analyzis, but the specifics matter, and if it is stable all is fine.


In general, instabilities will occur for a particular range of capacitances: too small, and they are beyond the unity gain limit, too large and the ratio of esr to reactance becomes larger for physical and mathematical reasons, damping possible instabilities.

Note that the output is not necessarily the best spot to detect some types of oscillations: if they are in the VHF range, the capacitor that causes them is going to attenuate them to the point of making them invisible with an oscilloscope and a regular X10 probe.
The output of the opamp would make a better test point.

Anyway, a step-test is always a valuable and revealing tool.




I did look at larger caps at 630V, but to my surprise 600V+ electrolytics are very rare.
It is a technological limit, linked to the properties of aluminum oxide.
The alternative, tantalum, has an even lower limit, so unless another miracle element is found, the 550V~600V limit is here to stay
 

PRR

Member
Joined 2003
Paid Member
> to my surprise 600V+ electrolytics are very rare.

As Elvee says: simple AlOx wants to leak bad at 450V. Fancy processing can get 500V rating.

There used to be 600V e-caps which were really two 350V caps in one cardboard tube. Not seen in decades.
 
AX tech editor
Joined 2002
Paid Member
> to my surprise 600V+ electrolytics are very rare.

As Elvee says: simple AlOx wants to leak bad at 450V. Fancy processing can get 500V rating.

There used to be 600V e-caps which were really two 350V caps in one cardboard tube. Not seen in decades.

Never knew that. I though they simple weren't in the catalogs because of lack of demand.
Maybe that is also the reason for those paper-in-oil high-voltage caps?

Jan
 
500V - 600V electrolytics are good, but we must move away from the usual sources, and especially away from « audio » types (some of which, like Elna Cerafine, are unreliable & fail short-circuit).

Parts of excellent quality can be had from the Kemet range that they acquired with BHC.

Even at the lowest cost range, the ALC10, offers 500V, 550V and 600V parts, and their leakage specification is actually better than ordinary electrolytics: 0.006CV (compared to the usual 0.01CV to 0.04CV).

Thee are superior performing parts, check out the data sheets. But do they sound any good? Well, some folks are upset by subjective comments, so I will sandbox them in a PGP stylee. Skip the next paragraph, if you have the allergy.

----- BEGIN SUBJECTIVE MESSAGE BLOCK -----

ALC10, ALS30 and especially ALS60 series sound better too, when replacing nasty consumer-grade 'lytics.

----- END SUBJECTIVE MESSAGE BLOCK -----

Some examples:

Low Cost ALC10
ALC10A121CC550 | KEMET 120μF Electrolytic Capacitor 550V dc Snap-In - ALC10A121CC550 | RS Components


High grade ALS60
leakage is 0.003 CV, lifetime 29000 hrs+ at 85 °C
ALS61A821KF550 | KEMET 820μF Electrolytic Capacitor 550V dc Screw Mount - ALS61A821KF550 | RS Components
 
Member
Joined 2004
Paid Member
I have several 60 microFarad MKP film capacitors rated for 900Vdc. Would it be beneficial to add one across the output of the regulator? I plan to use the regulator for the screen supply (only 200Vdc) of two channels of PP EL509 tubes that can draw a surprisingly large current on music peaks.
 
Hi Jan,


I just finished building your HV regulator (latest version with D3 zener 2.7V, with the board and components you sent me). With R11 being 220K I get an output voltage of 117 - 165 volt using the 5K VR pot, using a 3 mA min. load (resistor).


For R11 being 220K I would expect 127 Volt (using your formula Vout(max) : 0,58). However I get 165 Volt (max). How exact is your calculation formula?


So for 400 Volt out I would need R11 to be 689K?

Regards, Gerrit
 
Last edited:
AX tech editor
Joined 2002
Paid Member
I have several 60 microFarad MKP film capacitors rated for 900Vdc. Would it be beneficial to add one across the output of the regulator? I plan to use the regulator for the screen supply (only 200Vdc) of two channels of PP EL509 tubes that can draw a surprisingly large current on music peaks.

I would place these caps as close to the load, the screen, as possible. They help best there, the regulator can take care of itself ;-)

Jan