Feedback wanted: designing a power supply for electrostatic headphone tube amp

[mbrennwa]

The Open Source DHT Estat Headphone Amp (OSDEHA) needs a power supply!

The main thread about the OSDEHA development is in the Headphone Systems section (see here), but I don't think the power supply gurus will notice it over there. The purpose of this post here in the PSU section is to get feedback for the PSU design draft described here. I also attached the PSU and main amplifier schematics to this post, together with my breakdown of the PSU design notes.

Please shoot with comments and thoughts!

Breakdown of the different PSU sections (starting at the bottom of the schematic):

• The DC supply for the heaters of the input-stage tubes is nothing fancy. Just a full-wave rectifier followed by an LM1085 to get 6.3 VDC for the heaters. The heaters of the 6E5P tubes take up to 0.65 A each. There are two 6E5P per channel (four for stereo), so we need about 2.6 A in total. The Coleman filament regulators for the DHT filaments are on separate boards, so they are not shown here.
• On top of the heaters DC rides a delay circuit that turns on the B+ and B- supplies a few seconds after the heaters are up. The circuit is inspired by the delay used in the High-Amp AC amplifiere (see here). It's 555 timer with 1MΩ + 22uF RC element controls a relay that turn on the transformer secondaries of the B+ and B- supplies.
• The high-voltage supplies (see separate drawing) for the B+, B-, and buffers/grid-bias are inspired by a circuit shown by @mogliaa (see here). It is a relatively simple concept using a CCS for a stable voltage reference (2 x LND-150 cascode). The voltage drop across the "RSET" resistor is buffered by ZTX560 BJT and applied to the STF5NK100Z FET that outputs a smooth DC. There are many other schematics/concepts for HV PSUs out there, but they usually rely on a feedback mechanism for regulation of the output voltage. This may provide excellent measured performance, but a PSU without feedback regulation tends to provide better sound in my experience.
• The bias voltage for the headphone stator is inspired by the Golden Reference HV PSU (see here). It's basically a "charge pump" from the transformer secondary AC riding on the B+ DC output. When the AC swings negative and falls below the DC voltage derived from the B+, the B+ supply charges the capacitor connected to the AC secondary. Once the AC starts swinging positive again, the capacitor charge is pumped through the 10M90S and into the tail of the circuit, where the voltage is limited to 580 VDC by the Zener string.

 

[PretentiousFood]

I'm a litle concerned about your current limiting resistor, R30. With no feedback around it, its impedance adds to the regulator's output impedance. 33 ohms might not seem like much when driving a 100pF load, but I've found supply impedance to make a suprising difference in Stax amps. I'd also worry a bit about dissipation on the ZTX parts-- looks like just over a watt across the two devices? Q11 could likely be replaced by a depletion mode device and get you a bit more ripple rejection with fewer parts.

I've attached Gary Pimm / John Swenson's regulator. While this does use feedback, you could drop the error amp and use the same foldback current limiting (Q6) and get away with a smaller resistor in series with the source.

I used shunt regulators (which are inherently current-limited) in my ES amp. Attached the schematic for both rails. Since the CCS and shunt leg are both two-terminal devices, you can use N-channel parts with the CCS in series with the rail, and connect the middle point to ground.

 

[mbrennwa]

All very good points! Thank you!

I modified the HV PSU as below (see also attachment). Let me know what you think.

Q11 could likely be replaced by a depletion mode device and get you a bit more ripple rejection with fewer parts.

Ok, replaced this with another LND150 CCS. I will have to check the 147R value though.
Do I need the Zeners in this position?

I'd also worry a bit about dissipation on the ZTX parts-- looks like just over a watt across the two devices?

Oups, yes! I added a power resistor to the collector of the lower ZTX to burn some of the heat. I chose the value such that the 3 mA CCS (LND-150 with 147R resistor) never sees (much) more than 200 V.

I'm a litle concerned about your current limiting resistor, R30. With no feedback around it, its impedance adds to the regulator's output impedance. 33 ohms might not seem like much when driving a 100pF load, but I've found supply impedance to make a suprising difference in Stax amps. ...
...I've attached Gary Pimm / John Swenson's regulator. While this does use feedback, you could drop the error amp and use the same foldback current limiting (Q6) and get away with a smaller resistor in series with the source.

Current limiter changed as in the Pimm / Swenson example.
Did I get this right?

 

[PretentiousFood]

Ok, replaced this with another LND150 CCS. I will have to check the 147R value though.
Do I need the Zeners in this position?

I've never bothered with zeners on depletion mode CCSs in general. I know some folks have had bad luck with DN2540s, but my sense is that with G connected to S, there's not much opportunity to exceed the datasheet values (compared to, say, a follower).

Also, the LND150's transconductance is low enough that I don't bother with a gate stopper anymore. 🙂 I haven't had any trouble with them oscillating.

Oups, yes! I added a power resistor to the collector of the lower ZTX to burn some of the heat. I chose the value such that the 3 mA CCS (LND-150 with 147R resistor) never sees (much) more than 200 V.

The series R increases the follower output Z a bit. I don't have anything better to sugest as there aren't many 500V PNPs/P-channel devices to work with, but it's something to be aware of.

The LND150 has an IDSS of minimum of 1mA. In practice, I've never seen any higher than 2.5mA. If you do need the full 3mA, you might be better served by something like the BSS126 or with two LND150s in parallel. You could cascode it or add a resistor in series with the drain to keep heat off the package.

Current limiter changed as in the Pimm / Swenson example.
Did I get this right?

That works. What you drew there is a simple current limiter whereas Pimm uses a foldback limiter. I don't think it matters which you use in an amp like this, but here's some more info to help you decide.

It might make sense to add a jumper to bypass the R once the amp is all boxed up. I do think even 2R is enough to make a difference, but a jumper will make it easy for you to decide for yourself. 🙂

I don't see any spots for a PS bypass cap. Are there any spots on the amp board?

 

[mbrennwa]

Thank you again for your super useful insights! See revisions in the attachment and comments below.

I've never bothered with zeners on depletion mode CCSs in general. I know some folks have had bad luck with DN2540s, but my sense is that with G connected to S, there's not much opportunity to exceed the datasheet values (compared to, say, a follower).

Good point, as always. I ditched the Zeners from the CCSs in the PSU.
Will also ditch them from the CCSs on the amp boards.

The series R increases the follower output Z a bit. I don't have anything better to sugest as there aren't many 500V PNPs/P-channel devices to work with, but it's something to be aware of.

My thought was that the current through the resistor is fixed by the CCS, so the collector would ride on a fixed voltage. Am I wrong?

The LND150 has an IDSS of minimum of 1mA. In practice, I've never seen any higher than 2.5mA. If you do need the full 3mA, you might be better served by something like the BSS126 or with two LND150s in parallel. You could cascode it or add a resistor in series with the drain to keep heat off the package.

Ahhrgh, I looked at the Vgs chart in the datasheet upside down.
The BBS126 does not come with a through hole package, so I'd like to stay away from that.
I don't think I need 3 mA, so I could just run the LND150 at its IDSS with the gate connected to the source (Vgs=0). If I am wrong and it would be better to use more current than the IDSS from a single LND150, I'll use two in parallel.

...but here's some more info to help you decide.

ESP is just great, as always!

I don't see any spots for a PS bypass cap. Are there any spots on the amp board?

Yep, I'll add 50 uF caps to B+, B- and the buffer/grid-bias supplies on the amp boards.

 

[PretentiousFood]

My thought was that the current through the resistor is fixed by the CCS, so the collector would ride on a fixed voltage. Am I wrong?

No, you're right. However, the impedance is reflected back to the emitter by some relation that I am not clever enough to figure out. It's not unlike building an output stage that runs off of a high impedance power supply.

I attached a Spice sim. Not sure how accurate the models are, but the follower with no collector resistance has an impedance of 17 ohms, which aligns well with theory (1/gm = Vt/Iq = 26mV/1.5mA = 17R). The one with the the series R is about double, and, depending on how much you trust a sim, might get weird above 100kHz. Bypassing it helps, but introduces an extra time constant and might give strange startup behavior. The rightmost example burns the voltage across some 100V zeners. I think this is how I'd do it?

On the other hand, does the follower bring anything to the party?

I don't think I need 3 mA, so I could just run the LND150 at its IDSS with the gate connected to the source (Vgs=0). If I am wrong and it would be better to use more current than the IDSS from a single LND150, I'll use two in parallel.

I'm not too sure how much current you need, but I can't imagine the gate of a MOSFET pushing no more than 300mA would need much. Hopefully someone else can chime in.

That said, I'll suggest replacing the STF5NK100Z with something that has lower capacitances. SiC is getting sort of affordable and the TO247 package is easy to isolate as you don't need a shoulder washer, just a ceramic insualtor and regular bolt. Maybe NTHL1000N170M1? On the cheap end, I've had good luck with the Toshiba TK4K1A60F, but it's only a 600V part. Might be cutting things a bit close?

Yep, I'll add 50 uF caps to B+, B- and the buffer/grid-bias supplies on the amp boards.

I was thinking more like a 0.1uF film cap? Your regulator will have a 2-4 ohm output impedance, so be aware that hanging 50uF off of that will create a time constant in the 800Hz to 1.6kHz range. I'd try to keep it well above 20kHz, i.e. 4uF max.

 

[mbrennwa]

That said, I'll suggest replacing the STF5NK100Z with something that has lower capacitances. SiC is getting sort of affordable and the TO247 package is easy to isolate as you don't need a shoulder washer, just a ceramic insualtor and regular bolt. Maybe NTHL1000N170M1?

Nice! I simply copied the STF5NK100Z from @mogliaa's design from 10 years ago, and I guess these low capacitance parts were not available back then.

I attached a Spice sim. ... The rightmost example burns the voltage across some 100V zeners. I think this is how I'd do it?

On the other hand, does the follower bring anything to the party?

I like your Zener idea, except that Zeners tend to be noisy.
The question about the purpose of the follower is a good one. I simply copied the schematic from @mogliaa, as described in the first post without thinking enough about it. Looking at the schematic I agree that the gate of the output FET could be driven driven directly off the voltage reference of the LND150 cascode CCS. Maybe the idea of the follower was to drive the capacitance of the FET gate to improve the high-frequency bandwidth -- but I don't know. I am tempted to remove the follower and make things simpler, but hopefully not too simple.
It would be interesting to simulate the whole thing to see the effects of the follower (if any), but I am 100% clueless with SPICE and stuff.

I was thinking more like a 0.1uF film cap? Your regulator will have a 2-4 ohm output impedance, so be aware that hanging 50uF off of that will create a time constant in the 800Hz to 1.6kHz range. I'd try to keep it well above 20kHz, i.e. 4uF max.

Oups, yes!
So you'd not use any "big" caps to filter the output from the high-voltage supplies at all? I guess one could argue that the whole amp works in balanced Class-A, so large current spikes are not expected. I am more than happy to avoid any electrolytic that takes up space on the boards and will dry up after a while.
After building a bunch of Pass loudspeaker amps it feels awkward to not have many big caps in there...

 

[PretentiousFood]

I like your Zener idea, except that Zeners tend to be noisy.

I'm not sure that this is actually true. So I took some measurements with a BSS126 1.75mA CCS.

The noise spikes are likely picked up by the high impedances of the measurement rig-- I wouldn't look too closely at them.

2uV/sqrt(Hz) seems like a lot, but consider that this is on a 100V reference-- making them only slightly worse than 60 or so LEDs in series. At low frequencies, below 50Hz, the resistor is noisier than the the two active devices, perhaps because the current source's 1/f noise is converted to a voltage noise by the resistor. At very low frequencies, bypass caps can become impractically big, but this sort of noise is likely not audible anyway.

I think you could make the case for using any of these options, depending on whether you're optimizing for broadband noise, low frequency noise, or number of purple tubes. 🙂 Ultimately, they really don't seem that different. Be mindful that any of these will drift, too.

So you'd not use any "big" caps to filter the output from the high-voltage supplies at all? I guess one could argue that the whole amp works in balanced Class-A, so large current spikes are not expected. I am more than happy to avoid any electrolytic that takes up space on the boards and will dry up after a while.

I keep the big caps before the regulator. Indeed, there should be no large current spikes-- your output stage can't swing more than 40mA anyway. Before the regulator, I'd suggest a couple of 400V electrolytics with a ballast/bleeder resistor across each to balance the voltages. Do plan for a physically large film cap after the regulator, though. It would be nice to run one from B+ to B- in addition to a cap from B+ and B- to ground, since the output stage spans the two rails.

Also, be prepared to check that fancy SiC part for oscillation. You'll likely need to do some work dialing in the gate stopper.

 

[mbrennwa]

So I took some measurements with a BSS126 1.75mA CCS. ...

Whoa! Ok, the Zener is a nice option -- assuming the follower is needed. Do you think it is?

I keep the big caps before the regulator. Indeed, there should be no large current spikes-- your output stage can't swing more than 40mA anyway. Before the regulator, I'd suggest a couple of 400V electrolytics with a ballast/bleeder resistor across each to balance the voltages.

Hmm, anything wrong with a single cap with a higher voltage rating?
(Yes, bleeders are in order.)

Also, be prepared to check that fancy SiC part for oscillation. You'll likely need to do some work dialing in the gate stopper.

Ok, thanks for the heads up. How do I check for oscillation?