Solid state high voltage audio frequency amplifier

[Marco]

I was wondering what the options were for a solid state amplifier for small electrostatic speakers and I stumbled across this design :
https://www.edn.com/class-ab-inverting-amp-uses-two-floating-amplifier-cells/
I like the concept of using two floating N-channel based voltage to current converters as amplifier cells, but from eyeballing it the temperature stability, quiescent power and bandwidth of the particular implementation seem to me lacking. It doesn't look good in simulation either.

Would the upper circuit in the following picture work in practice for an amplifier cell? With a single MOSFET instead of a stack, because stacked transistors are a headache. Would need to find some ancient planar MOSFETs which don't self destruct when used linearly.

Graph is the current the two circuits can drive as close to clipping as possible, simulated for 20 degrees temperature variation. The Sziklai cascode seems pretty much superior in every possible way (didn't simulate the full stack for the cell from the article, but that doesn't matter).

[Kleinstein]

One can use cascaded transistors of MOSFETs. So the voltage is divided over 2 or 3 stacked transistor. This is used in most higher voltage SMUs. It also helps to have a few supply rails, not just to reduce the power loss, but also to reduce the worst case voltage with a reactive load.

[Vovk_Z]

Here in an attachment is a simplified version of that cascode HV-amp.
I used OPA37 with it. It only needs to adjust R34/R35. It works fine for low and middle audio frequencies but it doesn't have a wide frequency response.

[Marco]

Here in an attachment is a simplified version of that cascode HV-amp.

That's not a cascode, that's an emitter follower driving the MOSFET. I'm interested in an actual cascode because all the real current is drawn from the MOSFET drain, the bias resistor from the rails can be huge (10 Meg in the toy example I posted).

Also better thermal stability allowing you to set the quiescent current of the cells lower. Current transfer from LED to photodiode, current multiplication by the optcoupler BJT beta, current multiplication by the Sziklai PNP (plus diode). It's all multiplicative, no significant offsets. The gain of a stage might change a little, but that has far less influence than offset voltage drift in a voltage driven amplifier stage.

[Marco]

One can use cascaded transistors of MOSFETs.

You can, but if you do it with a resistor divider you have to compromise between quiescent power in the resistor divider and bandwidth. You could have an isolated driver stage for every MOSFET in the stack, but that's a lot of complexity. Stacked transistors are a headache.

[TimFox]

It may not be fashionable these days, but the two-opto-couplers to two MOSFET amplifier above is reminiscent of older audio amplifiers employing only NPN transistors that used transformer coupling to the two output devices.

[CaptDon]

I thought the impedance of electrostatic speakers was so high that you won't be passing much current? Which mandates the need for high voltage drive to produce any real audio power. Maybe a big heatsink will cure most of the ills?

[TimFox]

The capacitance of full-range electrostatic speakers can be quite high, so they need current at high frequency.
(They can be far from 8 ohms).
A typical value could be 2 nF, but usually they are driven through a step-up transformer (maybe 100:1 voltage ratio) which presents a much higher capacitance to the (low-Z) terminals.
I run a pair of "hybrid" speakers (dynamic woofer, crossing over to the ESLs at about 500 Hz), which do not tax my nominal 60 W/channel into 8 ohms solid-state AB amplifier.
Purpose-built tube amplifiers for ESLs drove them directly from push-pull plates.

[Marco]

This would be mostly a toy, to get real power you'd probably need to parallel MOSFETs ... or switch to tubes.

There are a handful of high voltage MOSFETs and IGBTs with decent DC SOAs, but I'm not sure if they aren't just complete and utter bullshit (even with derating for case temperature those are some pretty incredible specs).

[T3sl4co1l]

Yeah, ESLs need quite high voltage.  The impedance is somewhat capacitive (as you'd expect, given the structure), but it's also quite low Q -- the electrical efficiency is high -- so the impedance changes fairly slowly with frequency.  At least, that's my understanding; I don't have any to measure, but a friend works with them regularly and has said as much.

So, it's just the usual impedance matching problem.  Tubes are fine because they're well suited to kohm range impedances.  Transformers are fine, if a bit obnoxious to get into the same range (bandwidth suffers -- but not much is needed for audio, so not much complexity is needed in the transformer design, usually a couple sections interleaved will do).  Semiconductors aren't great because of the voltage range, and you need to cascode stacks and stuff.

There are MOSFETs available rated 2.5kV and up, and I think that includes "linear mode" types.  (I haven't checked in a while, I may be misremembering this.)  These would be fine in an amp, you just need to arrange something for the high side gate drive -- low side can be negative supply referenced just fine.  Or, well, could just as well call the high side source common and use a floating supply, doesn't really matter, but yeh either way one of the two needs quite a level shift, or isolation, to get there.

I don't particularly like the opto approach above, but maybe it's not too bad with some work.  If it can be calibrated or compensated in some way, or put in a feedback loop say to servo high and low side currents, then the gain error can be factored out.  For sure, avoid the emitter follower wiring: this specifically cancels out the one advantage of the 6N136, its low Miller effect from the independent photodiode.  But that's about it; I do strongly recommend SFH6345 or others instead -- the base pin of 6N136 means it's essentially useless for any kind of CMRR/immunity -- take careful note of what voltage and rate it's tested at!

I once drew a circuit with a similar sort of function, in terms of servoing -- using local gate drive amps with feedback to set output currents.
https://www.seventransistorlabs.com/tmoranwms/Circuits_2010/Power_OTA.png
Note that Vbb's may need to be isolated, but a bootstrap is also possible.  The current mirrors should probably be improved, given how much voltage/swing some need to handle (a cascoded or Wilson style would be fine), and a bias resistor instead of the pullup mirror in the gate amps would do (assuming Vbb's are fairly stable; which might not be so much the case in a bootstrapped design, so consider that as well).  But yeh, you can see the general scheme, it's currents into currents into currents and the output simply follows from that.  The major downside with this design, as I recall when I was simulating it (I never actually built it, alas), the bandwidth is quite high, some MHz -- but the phase shift is also quite large up there, so it can't run closed-loop up there.  Well, if your main bandwidth limitation is the load (say because it's capacitive -- and maybe adding some R+C and C additionally to compensate for cable inductance and whatnot), that should solve that, at least in part, and then you've just got a much faster transconductance amp driving a slow cap, servoing its voltage via negative feedback, which is a very easily compensated arrangement.


Aside, these HV MOSFETs are also available in far higher current ratings (lower Rds(on)) and speed than tubes, so even in a heavily biased comparison to the best sweep tubes (i.e. ignoring screen and heater power), transistors win on all accounts.  Earlier, HV MOS were rare to see, if available at all; I think the first was a 150 ohm 2500V part IXYS introduced some time in the '00s or '10s, which compares favorably with sweep tubes like EL519 (6500V, a bit over an ampere, "Rak(on)" ~ 30 ohms), but doesn't quite win on all accounts.  Parts up to 4kV are available now, and with Rds(on) < 10 ohms too I think, so there's basically no matchup anymore.

So yeah, not a big deal, just any other audio amp, scale it up for the required voltage and that'll do.  A shame that high voltage VAS types aren't available; maybe one of those smaller (<1A) HV MOSFETs would do, but you still don't have complements, so like noted above, circuits tend to be designed very much like the old school all-NPN (or all-PNP if you want to go really far back, heh) designs.  AC coupling is your friend; this is audio after all, and ESLs only go so low anyway -- consider using AC coupling, bootstrapping ("Q capacitor" from output to VAS pullup resistor: keeps ~constant pull-up current within the design bandwidth), etc.  And anyway, if your alternative is a transformer, AC coupling really isn't costing you anything in performance, comparatively.

Tim

[strawberry]

Linear isolated circuit

[David Hess]

I would try a direct coupled bootstrapped design, but I would worry about short circuit protection.

https://www.edn.com/bootstrapping-your-op-amp-yields-wide-voltage-swings/

[Conrad Hoffman]

IMO, it's hard to do if you look at the slew rate needed based on voltage and bandwidth. Then you have the capacitance to drive at high frequencies. Makes me want to use a lower voltage amp and a transformer!

[Marco]

IMO, it's hard to do if you look at the slew rate needed based on voltage and bandwidth.

The cascode can drive the HV transistor pretty hard AFAICS, plus ~10V and minus whatever should be enough.

Anyone have any reason to believe the Sziklai cascode wouldn't work as simulation predicts other than current crowding blowing up the high voltage transistor?

[magic]

There is no Sziklai-anything here. You have a current mirror multiplying the optocoupler current ~11x and pulling it from the FET's source.

Common-gate drive of the FET looks like a very good idea, because it sidesteps all problems with Vgs transfer nonlinearity and thermal drift, which would be a PITA for common source control. Stage current is controlled by small signal circuitry, which can be kept cool and made accurate. Mirror linearity might be improved by replacing D1,Q1 with 11 identical PNPs. Or heck, even an opamp. The optocoupler would probably prefer to operate near its peak CTR for minimum CTR variation with signal. Increasing output gain would reduce opto current variation and CTR modulation, but make output bias more sensitive to CTR drift.

There is a potential concern about gate currents flowing through the circuit and introducing unaccounted nonlinearity.

edit
Dunno about stability and magic-smokability. MOSFETs like to oscillate at tens of MHz in linear mode given any excuse, but here C1 bootstraps the small signal circuit to gate voltage, so its supply voltage will not vary a lot and hopefully it will hold its constant current firmly. FWIW, depletion FETs with source-gate resistors are used as current regulators quite often.

[Marco]

There is no Sziklai-anything here. You have a current mirror multiplying the optocoupler current ~11x and pulling it from the FET's source

A PNP current multiplier added to a NPN transistor is a Sziklai composite transistor. This one just has emitter degeneration.

Mirror linearity might be improved by replacing D1,Q1 with 11 identical PNPs.

Absolute linearity wasn't really the aim for adding the diode, just wanted the two transistors turn on around the same time, minimize the zero crossing cleanup for the outer loop. Linear enough, though 11 transistors is an option.

The optocoupler would probably prefer to operate near its peak CTR for minimum CTR variation with signal.

Probably, but keeping the entire cell a current multiplier makes biasing really easy. Intention was to let the outer loop deal with the reduced current gain at low current.

[strawberry]

IMO, it's hard to do if you look at the slew rate needed based on voltage and bandwidth. Then you have the capacitance to drive at high frequencies. Makes me want to use a lower voltage amp and a transformer!

general purpose transformer will cut off at ~10kHz. due to skin effect, core...
Tube Audio output transformers.

[magic]

A PNP current multiplier added to a NPN transistor is a Sziklai composite transistor. This one just has emitter degeneration.

OK, I suppose you could say that, although I wouldn't because it isn't exactly the usual configuration. It doesn't have β² current gain, for starters.

Absolute linearity wasn't really the aim for adding the diode, just wanted the two transistors turn on around the same time, minimize the zero crossing cleanup for the outer loop. Linear enough, though 11 transistors is an option.

I thought about linearity because optocouplers may not permit a lot of global loop GBW to straighten things out.

Probably, but keeping the entire cell a current multiplier makes biasing really easy.

Can't disagree with that, though

[profdc9]

How about something like the circuit below, but with a high voltage cascode stack substituted for Q5, Q3/Q6, and Q1/Q2?  This uses MJE340/MJE350, but with a cascode of, for example, MJL21193/MJL21194 probably quite a bit of output power could be dissipated per device, especially with a high voltage stack for Q1/Q2.

There are single NPNs that go up to 1 kV VCE.  If you flipped all the transistors NPN to PNP in the below design, Q5 is NPN, Q3 is NPN, Q4 is NPN, Q2 is NPN, and Q1 is PNP, and Q6 is PNP.  Since there's no high voltage PNPs available, you could make it class A with Q1 being a resistor, and Q6 could be stacked low voltage PNPs.

[T3sl4co1l]

Yikes, double suicide biased BJTs?  BJTs at high voltage, linear?  Double yikes!

But yes, the basic idea would be fine; emitter degeneration on Q5 and Q6 makes it a simple transconductance level shifter, and the CCS pull-up gives it high voltage gain (mind, way, way too much gain to close the loop on an op-amp).

Q3-Q4 is problematic especially at high voltages, where Q3 runs away due to raw unbridled Early effect plus self-heating.  A cascode is mandatory.  For higher voltages, a divider chain between Vcc-Out-Vee (wait why is that "-Vcc" and not "Vee"?..) will be needed; note it doesn't need to be common to the gain node, it can be bootstrapped.  One advantage of the follower topology.  Which means the outputs themselves can be cascoded on, well it can be the same divider chain really, can't it.  So that's not too bad.

I don't get R4 and R5.  Something about frequency compensation perhaps (Ccb of Q3, Q6)?  (Surely it's not current limiting, a triple-suicide-biased route?)

That said, some resistance in strategic places might go some way to improving robustness, i.e. in event of output short circuit or overload (or teething troubles while testing ).  Consider what happens when any given transistor fails shorted, and ensure adequate resistance around to prevent overvolting the remaining parts.  Might give some chance of turning it off before the whole thing runs away.

Output also shows class C, which is pretty nasty for audio, even at these voltages.  Easily solved with a couple bias diodes, which can be integrated into the cascode divider chain as well, I think.

Can also be made complementary, by mirroring Q5; in that case, both Q6 and Q3-Q4 act as current mirrors; note that Q6 can have more than unity gain, unlike the plain mirror.  So that's probably a worthy improvement.

Minimizing voltage swings is mandatory for bandwidth purposes.  Consider Q5: it could be turned into a cascode chain, with fixed base voltages, and minimal emitter degeneration (I mean, enough to handle the op-amp's output swing, but little in the stack itself, and a low output (collector) swing (i.e., at R7).  As opposed to using series collector resistors (kinda like R4 and R5), which introduce a 1/(R*Ccb) pole, and Ccb is of great importance when so little bias current is available (to minimize power dissipation in these level shifting and driver sections).

Speaking of current, Darlington outputs are generally undesirable for their higher saturation voltage, or perhaps lower bandwidth; but they are probably an excellent idea here, biased properly of course (use some Rbe across the power transistor to bias the driver into reasonable conduction).  Who cares about saturation when you have thousands to spare; meanwhile, the higher composite hFE minimizes drive current (which will still be limited by node capacitance and required bandwidth, but the Darlingtons help with this in part, too).

Anyway, that's all assuming BJTs that actually...work, which is a big ask at high voltages.  As I recall, there's basically nothing above 400V or so, for linear operation.  BJTs are just terrible at voltage, in linear operation.  In particular, you can't find any amplifier transistors over 250V or so; all 1.5kV (or higher) parts are optimized for switching, with embarrassing SOAs.  Anyway, MOSFETs basically drop into most of these positions, with the higher Vgs(th) vs Vbe being irrelevant, and the low Ig a great improvement (assuming, of course, Ciss is manageable, sometimes a hard ask when forced to choose among power switching types).  Do include a S-G zener to prevent gate rupture; make sure it's a small part (low capacitance!).  BJTs can still be used at the bottoms of cascodes if you like, for higher transconductance or better current matching or what have you; indeed this can be designed to quite a low voltage drop so that say 2N3904/6s do the job, since we're not talking a whole lot of current here in any position.

Ed: Mind, this isn't personal.  (I think. I hope?)  Note the filename of the above attachment... one should never be surprised what kind of trash may be found on those ezines.

Tim

[magic]

I have never used such things at 1kV but it would probably still work: an HV depletion FET (if you can find a suitable one) with a source-gate resistor could act as active load in absence of P polarity devices.

See TLC272 for a simple VAS and push-pull output stage using only N-channel and one active load, which could be as above or perhaps the oldschool bootstrapped resistor.

The problem of transferring control signals to the negative rail may be eliminated by using a single HV supply and AC coupling the load.

That current mirror with two power transistors is absolutely laughable. All the matching is in vain since it will never work for reasons given above, might as well replace the left one with 1N4148 after adding the mandatory degeneration resistors.

I think all those 10kΩ are for base current limiting
One is still missing right on the output of the opamp

If you fix the mirror, protect Q5 and make sure that all limiting resistor values are right, the power oscillator will be good to go

[cdev]

I am not volunteering my devices, but for years I have had two IC's they are PA-08 by Apex, I think. They appear to be broadband op-amps that operate at high voltages. Ive been holding on to them for use with servos or mechanical devices that require high voltage and rapid response.
They work from DC to over 1 MHz, and support quite a bit of power if adequately heat sinked.

They look interesting. I got them at an estate sale, with sockets. They were owned by a retired engineer for RCA, who appears to have designed solid state devices and knew a lot about transistors, ICs etc.

Maybe that device, PA-08 might suit your bill. I am curious about what I could do with mine.

[NiHaoMike]

What about a pair of solid state high voltage switches and a pair of high voltage DC/DC converters modulated with the audio signal?

[mawyatt]

I am not volunteering my devices, but for years I have had two IC's they are PA-08 by Apex, I think. They appear to be broadband op-amps that operate at high voltages. Ive been holding on to them for use with servos or mechanical devices that require high voltage and rapid response.
They work from DC to over 1 MHz, and support quite a bit of power if adequately heat sinked.

They look interesting. I got them at an estate sale, with sockets. They were owned by a retired engineer for RCA, who appears to have designed solid state devices and knew a lot about transistors, ICs etc.

Maybe that device, PA-08 might suit your bill. I am curious about what I could do with mine.

Last year we evaluated the Apex PA443 HV Dual amp when we needed outputs to +-150V (they are rated to +-175V). Quickly designed a built a test PCB (see below) to interface with an AWG, hosed up the PA443 footprint tho

The idea was to evaluate the amps performance with special type waveforms for a new Phased Array Controller which required 128 independent programmable channels capable of suppling +-150V precision waveforms. Additional research into the phase array voltage controlled elements reveled the voltage requirement could be relaxed to +-100V, so we ended up selecting the OPA462 for various reasons including footprint (we used 128 individual amps, later possibly expanding to 256 or even more).

Anyway, recall the PA443 performed well and had no serious issues. If one can live with the 350V (+-175V) max supply level, these might be a consideration if you can get them!! Edit: The data sheet shows (Fig 16) an application capable of supplying 660Vpp using a Push-Pull arrangement.

https://www.apexanalog.com/resources/products/pa441u-pa443u.pdf

Best,

[Terry Bites]

 Check out this page from tubecad, if you havent already. https://www.tubecad.com/2018/01/blog0407.htm A monster cascode https://www.tubecad.com/2018/01/01/Class-A%20or%20Class-AB%20Push-Pull%20Amplifier%20for%20Driving%20Electrostatic%20Speakers%20Large.png
PSU design and protection needs a good think through.

A UV lamp inside the speaker will keep the mosquitos under control. They keep the house dust down as an added bonus.