High Voltage Regulators (Maida or zener)

[tomchr]

Folks,

I figured I should let you know where I'm at with the prototyping... Stare at attached.
Note that C101, C102, C103, R107, C201, C202, C203, R207 have multiple options. The idea is to allow for either a film cap or an electrolytic cap in case of the capacitors. The resistor options allow for using 3W, 5W, or 10W types. This allows for a fair amount of flexibility as far as input and output voltages go.

R200 is in series with the input to the negative regulator. It ensures that the cascode device, Q201, doesn't fry in the event of a short circuit on the negative supply.

The image shows two boards on one panel. The end result will be two separate boards.

I plan to build a prototype using toner transfer before committing to a PCB fab run. I'll let you know how it goes.

~Tom

[Rod Coleman]

Originally Posted by tomchr

I disagree, actually. Run a sim of the loop gain and it's pretty easy to convince yourself of the advantages of the IC regulators. You do have to ensure that the cascode device will operate within its SOA under all operating conditions. If the cascode fries, so will the IC.

It's the same level of reliability for the discrete regulator. If the pass device in the discrete regulator fries due to operation outside its SOA, the regulator won't regulate.

~Tom

Running a sim closed-loop may hide some of the real problems.

If you try it open loop you can see the effect of the two poles formed within the stack.

The first, formed by the darlington input and the large base resistor, is at 20 or so kHz.

The second is formed by the current limit resistor plus the dynamic impedance of the 317 output transistor [ a few ohms at likely currents], against the output & load capacitance.

Provided you use all the values specified by Maida AND your load is broadly resistive, the transient response is reasonable. But for instance, increasing the load capacitance causes the second pole to fall lower down the audio spectrum, so that the reaction with the first pole gives a phase response going up & down like a *****'s drawers across your load spectrum - hardly conducive to a neat transient behaviour.

This is the reason that Maida specified the output network of 2.7R and 1uF MKT [or 1R and 1u elko if you cheap out], where the ordinary 317 circuit is quite happy with big capacitor loads. In a Maida, applying the capacitive load without the resistor would probably induce oscillation, for many C values.

Anyway - my point is that for a general-purpose regulator, where you don't know the nature of the load, and you might want to add capacitors anyway - it is better to have a regulator with a fairly low open-loop impedance. Getting plenty of loop gain is no problem with some discrete transistors, as Elvee has shown.

[tomchr]

Rod,

I see your point and do agree that open-loop sims are needed to reveal the poles and zeros of the loop. I think I misunderstood your comments about reliability which put me off for a bit. Anyway... I think your concerns are valid.

The prototype I threw together from the schematic/pcb in my post of yesterday actually works really well. It regulates fine with no load. However, it does not survive start-up into a low-impedance load. On the lab bench I use light bulbs for a load. Four 25 W, 120 V bulbs in series present 200-some ohms with cold filaments. If I bring the input voltage up slowly, the regulator works fine. But turn-on by the flip of a switch leaves me with $7 of dead silicon... Relying on the emitter/source resistor of the cascode to provide any current limiting function is not reliable. I may have to swallow my pride and use a discrete regulator...

On the upshot, this means I could get to have some fun by using a 0C3 or similar as the reference...

The need for the regulator to start up into, basically, a dead short is due to the Loftin-White circuit. The regulators "sees" a discharged 47 uF cap in series with a 600-ish ohm resistor on start-up. The regulator will need to charge the cap to about 400 V without dying. After that, the load current is 90~100 mA. SOA limit on the pass device I'm using is 300 mA at Vds = 600 V, Tcase = 25 C.

~Tom

[tomchr]

Attached shows my latest incarnation of the high voltage regulator. It's pretty straight forward. I chose MOS to avoid base currents of BJTs. Power BJTs tend to have rather crappy beta, hence draw an unfair amount of base current. This forces resistors to be lower, hence burn more power. I have a small-signal BJT handling current limiting by sensing the voltage across a 5.6 ohm resistor. The current limit is about 130 mA.

I still need to add the compensation network so this circuit is by no means complete.

However, as an engineer and, thus, perfectionist there is one thing that bugs me a little bit. Due to the Vgs variation versus temperature of the error amp MOSFET, the output voltage varies about 1 % across temperature (0 C --> 125 C). I would like to straighten that out a bit. Does any of you have ideas on how to do this?

The temperature of the error amp device and temperature compensating device need to track. So they could be two identical devices mounted back-to-back on the same heat sink. Easy. One could be hooked as a diode in series with the lower resistor (R2) in the feedback network. But in order for this to work, the two devices will need to operate at the same current density, hence, I need to burn 5~6 mA in the feedback network. A bit rich for my blood...

Any other ideas?

~Tom

[SY]

I assume R3 is your dummy load?

A 1% thermal error is not exactly the end of the world, but you can use a diode in the reference string to compensate if your circuit is really that critical. But that 1% error isn't even close to being the performance limiting factor- your main problem will be poor regulation because of the low feedback. A speedup cap across R1 will help, but not nearly as much as increasing the reference voltage to something more like 100-200V.

If it were me and I didn't want to use a Maida (big "if"- they have always performed reliably and well when I've used them), I'd have the error amp be a CCS-tailed diff amp so the the reference voltage could be filtered. I'd make the error amp with bipolars to improve regulation. You'll want to stopper the MOSFET gate- R6 is on the "wrong" side.

[tomchr]

Originally Posted by SY

I assume R3 is your dummy load?

Yep.

Originally Posted by SY

A 1% thermal error is not exactly the end of the world, but you can use a diode in the reference string to compensate if your circuit is really that critical.

That's certainly true. 4~5 V variation on 470 V is not the end of the world by any means. I can always adjust the voltage at hot temp as the devices will dissipate a fair amount of power and run hot anyway.

Originally Posted by SY

But that 1% error isn't even close to being the performance limiting factor- your main problem will be poor regulation because of the low feedback. A speedup cap across R1 will help, but not nearly as much as increasing the reference voltage to something more like 100-200V.

I haven't sim'ed regulation yet. But the loop gain is on the order of 60 dB so I would be surprised if the regulation was that bad. Not thereby saying it couldn't be improved. In fact I would like to improve it.

Originally Posted by SY

If it were me and I didn't want to use a Maida (big "if"- they have always performed reliably and well when I've used them), I'd have the error amp be a CCS-tailed diff amp so the the reference voltage could be filtered. I'd make the error amp with bipolars to improve regulation. You'll want to stopper the MOSFET gate- R6 is on the "wrong" side.

Correct about R6. Oops... Good catch.

I tried several incarnations of the Maida regulator. The main problem is that the current limiter is relying on the Vbe/Vgs of the cascode along with the drop-out voltage of the LM317 to work. For resistive loads or small capacitive loads, this is not an issue. But in my case, I'm driving 47 uF with the regulator so the regulator will need to be able to survive starting up into a dead short. Aside from changing the topology of the amp I'm supplying (which I have decided not to do), there's really no way around it. I like the Maida as well, but for this application, I'm afraid it's out.

I'd like to filter Vref as well. Diff pair... DUH! Why didn't I think of that.... That would also eliminate the thermal issues associated with the Vgs of the error amp in the circuit above.

I'll take another spin through the sim cycle...

Thanks for your inputs.

~Tom

[SY]

One thing that helps you is that the impedance doesn't have to be ultra-low. You're running a tube circuit and it's class A.

Have you looked at Jan Didden's HV regulator? Very nice design. I've used one on my breadboard with excellent results, and it will be in my next power amp.

[jackinnj]

Did you consider the HV Regulator from Horowitz and Hill? I modified one of my Heath HV supplies to use it.

[tomchr]

Originally Posted by jackinnj

Did you consider the HV Regulator from Horowitz and Hill? I modified one of my Heath HV supplies to use it.

The one with the op-amp as the error amp (2nd Ed, p 369, fig. 6.47)? Looks like a pretty neat circuit.

~Tom

[SY]

Something pretty similar done by Joe Curcio (again, I built a dozen or so and no field failures). He published one of his in the first issue of Glass Audio- it used a Maida for a reference, then an opamp on floating rails for the error amp. The versions I built (except one) had a reference voltage derived from a FET current source pushing though a resistor, then RC filtered.