300VDC linear regulator for tube amps

[Twospoons]

This circuit was born out of my desire for a linear regulator for the plate supply of a vacuum tube phono preamp.
Not being happy with a simple passive filter (I want the supply to be really quiet), this design is based on the old TL431 linear shunt regulator, with some extra bits to handle the higher voltages.

Note: this only exists in my computer so far - I haven't built hardware yet.

Bits of note:

• R3 simulates the amplifier load of about 30mA
• Q1 R1 ZD1 keep the voltage seen by U1 to about 7V
• R12 supplies a bit of bias current to keep U1 operating
• R9 R10 R11 provide a little feed-forward, which greatly improves line regulation around dc and improves low frequency performance
• Startup time is on the order of 2 seconds

If the simulations are correct ( and I usually find they're pretty damn good) this regulator has roughly 100dB of line rejection from 10Hz to 20kHz. Will be interesting to see if this holds up in real life, though I'm damned if I know how I'm going to test it.
I've included the SPICE file (for Simetrix) if anyone wants to have a play with this.
300Vreg.zip

[klugesmith]

Applause! I bet it'll be quieter than my old semiconductor-free 0 to 400 V regulated power supply.

I like your rendering of bridge rectifier front end.
Could an input choke reduce the ripple voltage at C4 usefully, or allow a less-derated power transformer?

In what frequency range are Zener diodes considered to be noise sources?  Does C1 address that?

[shrad]

I would simulate R3 as a sine current source and repeat it with different ranges of frequencies, just to be sure

Maybe the use of a high voltage mosfet could also be interesting?

Otherwise nice design, I've seen several and your is on the good side

[T3sl4co1l]

Suggest a resistor in series with C3, and a zener to protect U1 REF.  C2 can be much smaller.

May also want to add a series resistor between Q1 and U1, so you can add local compensation across U1.

R2 can probably be larger..?  Or maybe a Darlington isn't necessary.  Likewise D1 could be smaller (1N52xx family?) but that doesn't really matter.

The feedforward is neat, if it's in just the right amount, and phase; I wonder how much phase shift the REF pin takes on, though.  C3 affects it, as well as local compensation (if applicable).

And a current limiter, which is easy to add, and can even be foldback limiting to help save on worst case dissipation.

The big things are current limiting and C3, else the thing goes poof the first time you accidentally brush against 300V and short it to ground.

Tim

[Twospoons]

Thanks for your feedback everyone.

- yes c1 is there to reduce zener noise, but given the zener is not the reference for the 300V output it shouldn't be an issue anyway. Just for fun I'll have a look-see to see how much zener noise would appear on the output. UPDATE: the zener noise appears to be attenuated by over 100dB on the output. Not a problem.

- I did originally have a choke on the input but I didn't like the turn-on  surge it created, and given the performance of the linear reg, I decided to ditch it.

- I did  look at using MOSFETs, but the BJTs seem to work a bit better. Which is a pity as there are more options for HV MOSFETs than BJTs with decent Hfe.

- I haven't really checked the load transient response (which I should for completeness) , but given the nature of the load I'm not too worried - 99% of the load is just the quiescent current in the vacuum tube stages (its a small signal phono preamp). UPDATE: ran a 90% load step and the response is smooth as butter, no overshoot.

- Protecting U1 ref is a good suggestion. Will probably just use a diode from ref to D1 cathode. Or maybe a transil to ground. UPDATE: definitely a transil I think.

- C2 is that size because the same part is used elsewhere, and the first price break is at 10 pcs. The difference between theory and practice ....

- I originally had a single transistor as the pass device, but I was finding the dropout voltage to be too high for my liking. Reducing R2 helps but also pushes up the power dissipation in Q1. The darlington arrangement fixes that. R2 probably could be bigger now, but changing the Iq through Q1 and U1 has some curious effects on the rejection bandwidth. Maybe I'll investigate further.

- I know D1 is excessive - it just happened to be available in the simulator

- Current limiting is a nice idea and I think I could do it with one BJT and a resistor. Current limiting wont save the pass transistor  in the case of a short on the output (9W at 30mA). Foldback could but would be much more complex.  I think I will just have to accept the bang if I do something foolish.

- The feed-forward took quite a bit of tweaking to get right, and, as you say, is affected by the feedback network. Its not particularly sensitive though - I ran a MonteCarlo with 1% resistors and 10% capacitors to check.

-Additional compensation does not appear to be needed. I haven't seen any sign of instability yet.

[T3sl4co1l]

- I did  look at using MOSFETs, but the BJTs seem to work a bit better. Which is a pity as there are more options for HV MOSFETs than BJTs with decent Hfe.

The difference is in gate capacitance.  BJTs require less drive voltage (typically ~5x more voltage gain than a FET of equivalent ratings) and may have lower capacitance besides, so they are hard to beat for a lot of analog circuits.

To handle the gate capacitance, you could use the same traditional circuit, but instead of a Darlington, use a diode-biased complementary emitter follower (which can be low voltage, referenced to FET source), to consume less quiescent current but still offer full emitter-follower drive, compared to brute-forcing it with a pull-up resistor.

Can also consider a depletion MOS, so the gate drive voltage is below the source, rather than above.  The gate drive supply could be derived from a series zener diode, and this would also be a convenient place to insert the current limiter.

The error amp needs to be faster or higher-gain to account for the MOSFET's poorer small-signal performance, but that will be easy to achieve with the cascode configuration, and the reduced capacitance afforded by the driver.  Also a current source could be used for pull-up instead of a resistor, giving more dynamic range for the modest cost of a transistor and a few resistors.

Do include a zener or TVS from source to gate, don't want to pop that thing accidentally, especially under transient conditions.

- I originally had a single transistor as the pass device, but I was finding the dropout voltage to be too high for my liking. Reducing R2 helps but also pushes up the power dissipation in Q1. The darlington arrangement fixes that. R2 probably could be bigger now, but changing the Iq through Q1 and U1 has some curious effects on the rejection bandwidth. Maybe I'll investigate further.

...

-Additional compensation does not appear to be needed. I haven't seen any sign of instability yet.

*Gestures flamboyantly at the two paragraphs*

Probably it's close enough to fine that it hasn't been a problem, but it may be worthwhile just to put in the components in case they do become useful, even if they're usually populated with 0-ohm jumpers, or depopped.

Also, there's some difference between manufacturers, or their sim models at least; who knows which way it'll fall.

- Current limiting is a nice idea and I think I could do it with one BJT and a resistor. Current limiting wont save the pass transistor  in the case of a short on the output (9W at 30mA). Foldback could but would be much more complex.  I think I will just have to accept the bang if I do something foolish.

I mean, foldback is literally just a resistor divider from VIN to current limiting transistor base to VOUT.  Not really complex in its simplest form.    Something like actual constant-power limiting would suck, yeah, but, on that note -- you probably have enough time to apply thermal limiting, i.e. stick a thermistor on the divider as well, so the foldback collapses all the way as the thing rises over, say, 125°C.  (A chip thermistor, surrounded by heat dissipating pours, has a time constant of several seconds typically.  If you can show that, in those few seconds, the power transistors never exceed, say, Tj = 175°C or whatever, you're golden!)

Tim

[shrad]

current limiting may or may not be good with valves... in the case of a small preamp this is OK but if you add it on the same B+ transformer than a push-pull for instance, current variation will give global DV variation and gain/operation points of preamp will vary accordingly, preserving proportionality

I'll be interested if you elaborate on the mosfet and produce a working circuit, I'm into valves right now too but something specific that I'll document when a prototype will be running

[Uspring]

- The feed-forward took quite a bit of tweaking to get right, and, as you say, is affected by the feedback network. Its not particularly sensitive though - I ran a MonteCarlo with 1% resistors and 10% capacitors to check.

You should make the feed-forward adjustable. Its optimal setting depends on rather ill specified open loop gains and other factors. They might vary considerably between same type components.

[Twospoons]

This is where pragmatism sets in. The regulator only needs to be good enough, not perfect. I doubt the difference between 95dB and 100dB of rejection is going to be noticeable in the final  project.

Also a current source could be used for pull-up instead of a resistor, giving more dynamic range for the modest cost of a transistor and a few resistors.

I did try this (using an ideal current source) but it turns out the resistor is much better (by 20dB @ 50Hz)  - probably hidden feed-forward happening that works out to be in the correct phase.

you probably have enough time to apply thermal limiting, i.e. stick a thermistor on the divider as well, so the foldback collapses all the way as the thing rises over, say, 125°C.  (A chip thermistor, surrounded by heat dissipating pours, has a time constant of several seconds typically.  If you can show that, in those few seconds, the power transistors never exceed, say, Tj = 175°C or whatever, you're golden!)

Tim

Yeah, well thats the trick isn't it?  I have in fact done exactly that in a current loop driver with an unknown loop burden voltage, using a thermistor to reduce the supply V and limit power dissipation if the current driver got too hot.
But in this case the transistors come in a sot-223, and I  don't trust they would handle the thermal transient well.  A lightweight fuse after C2 might be a good option though - theres plenty of energy there to blow a fuse if the output got shorted.

[T3sl4co1l]

current limiting may or may not be good with valves... in the case of a small preamp this is OK but if you add it on the same B+ transformer than a push-pull for instance, current variation will give global DV variation and gain/operation points of preamp will vary accordingly, preserving proportionality

I'll be interested if you elaborate on the mosfet and produce a working circuit, I'm into valves right now too but something specific that I'll document when a prototype will be running

It's inside the loop, voltage remains stable until peaks exceed the limit.  One would hope the limit is set above the maximum peak demand of the load, whatever it is.

Tim