High voltage, low current output stage for class D amplifier

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I'm working in a class-D amplifier to drive a pair of (homebrew) electrostatic loudspeakers. My target voltage is something between 1.2kV and 2kV peak to peak. The amplifier will deliver less than 50 mA's in all circumstances and probably less than 1 mA in average. This means that Rds On is going to hurt little, even if it's in the low kilohm range.

There are two problems in this design: Level shifting and output devices. Adequate output devices should be relatively low-current components, as high current capability usually means high gate charge. As for level shifters, the options are to use either a digital level shifter or an optocoupler. However I'm a little confused in the device election so I would like to ask the wise people in the forum for advice in choosing these critical parts.

My idea is to see if it's possible to build a class D amplifier + electrostatic loudspeaker system demonstrating ultra high efficiency and low overall distortion (including the loudspeaker in the distortion computation), rather than building something that fits audiophile tastes, so a conventional half bridge amplifier using raw UcD topology seems fine.
 
Last edited:
output device parasitc C makes high V, high frequency, efficiency switching amps impractical, as do the limits of output filtering inductor parasitics with high V*T capability

audio frequency xfmr step up would be an easier option
 
Last edited:
The point here is that we are talking about switching a few watts only. Even with 5% efficiency this would fall within the SOA of a single output pair. The output inductor is not much of a concern, as inductors are damaged by excessive current, not voltage, and currents are going to be incredibly small in this design. The capacitor may be more of a concern, but nothing that can't be fixed by spending a few bucks in a HV cap and a larger inductor.

The big problem, as you noticed, is the output capacitance. My second-line defense against it is cascoding. It's an ugly solution for a class-d amp and phase delays will demand very careful attention, but it's a solution after all. However I believe there is some hope left about finding an adequate pair of transistors. The reason is that the voltage is not *that* much higher than in conventional amplifiers, only one order of magnitude, but current is three orders of magnitude smaller. Three orders of magnitude would mean decreasing the number of carriers in the FET by three orders of magnitude to have the same voltage drop, and the parasitic capacitances by a comparable amount. Moreover, since a voltage drop that would be awful for a 50V amplifier is irrelevant in a 1000V amp, we could even relax the on-state voltage requirement by an extra order magnitude. With such a relaxed conditions current-wise, it seems strange not to find a device that suits the design.
 
if you calculate circulating power for full V drive into the C load at the highest audio frequency you may end up with more than "a few watts only"

Class A push-pull can be 50% "efficient" - but you have to bias for 1/2 peak current at highest audio frequency into the panel's C

1/2 * 50 mA * 2kV = 50 W

IXTH03N400 Mosfets could do the job in a linear amp - if you can buy them

Class AB is possible, cutting power supply current, mosfet dissipation requirements


the IXYS devices could be used as switches too except for the slow body diode and the need for floating gate drive since complements aren't available at these V levels

I still have doubts about a efficient switching inductor design - mH inductance, kV insulation, spacing requirements, while keeping SRF several 10x the switching frequency min
 
if you calculate circulating power for full V drive into the C load at the highest audio frequency you may end up with more than "a few watts only"

Class A push-pull can be 50% "efficient" - but you have to bias for 1/2 peak current at highest audio frequency into the panel's C

1/2 * 50 mA * 2kV = 50 W

IXTH03N400 Mosfets could do the job in a linear amp - if you can buy them

Class AB is possible, cutting power supply current, mosfet dissipation requirements


the IXYS devices could be used as switches too except for the slow body diode and the need for floating gate drive since complements aren't available at these V levels

I still have doubts about a efficient switching inductor design - mH inductance, kV insulation, spacing requirements, while keeping SRF several 10x the switching frequency min

My 5% efficiency and "few watts" here relate to the power dissipated in the loudspeaker, not to the total amount of power flowing back and forth. These are -approximately- independent of frequency. The key idea here is that class D amps have very high efficiency with capacitive loads. Obviously there will be conduction losses, but it would take something like 5 KOhm Rds per device to waste 20% of that power.

With regards to the inductor, the total amount of energy stored should not be any higher than in conventional, low current high voltage designs. Thiner wire, more turns, same flux. Isolation is a problem but I don't believe it's a show stopper.

I will take a look at your devices. I'm also looking at vacuum tubes, but I believe there is some delay mechanism in these that I don't understand. Less than 1 pF of output capacitance is too good to be true, basically rendering the driver design trivial. There is also a part of me who wants to stay away from these faulty things because the entire heather wiring and unreliability of tubes would make debugging much longer in a device that poses an extraordinary danger to work with.

I understand this is an uncommon project and I may have to face many unforeseen complications, but there are things that work in its favor too. Low current means little EMI and less problems with parasitic inductances.
 
Last edited:
Have a look at KSC5042 bipolar transistors, although they are only 900V, so you'd need to bridge them.

The electrostatic loudspeaker design is already bridged so the voltage would be a bit on the low side, but the 2.8 pF output capacitance of this transistor is wonderful! I don't know about switching times for bjt's however.

Perhaps somebody could enlighten me about switching times in MOS vs BJT's vs Tubes.
 
The major drawback of bipolar transistors (when used for class D) is that there is a variable delay (increases with base current and temperature) from the instant base charge starts to be removed to the instant Vce starts to rise. This delay may range from 200ns in low current devices (500mA) to 2us in high current ones (30A).

When operating bipolar transistors at collector currents not much lower than the maximum rating (>1/4 max. Ic) there are also increased conduction losses during the first 200ns-2us of each switching period while charge is being stored in the B-E junction (which has to be removed later before turn-off can happen). Vce-sat starts high and approaches steady state value as more charge is stored (dynamic saturation process). The higher the base current, the more charge ends up stored.

Go for high voltage MOSFET, they excel at low currents.
 
Last edited:
The IXYS high voltage MOSFET I've just found seems to be the solution. Pity it's not in stock at Digi-Key.

Now the problem left is the level shifting. For this purpose I would like to use an optocoupler or digital level shifter, but I'm worried on the upper side of the totem pole, because the drain is referred to the (unfiltered) output and not to a constant voltage, and high frequency signal may travel along the level-shifter ground and leak trough the mains transformer.
 
May be for these small currents ionomolo can adjust a filter ripple current which is higher than the max required output current.
This would avoid any currents through the body diode.
The output stage would always see an inductive load.
For such high voltage devices this would probably be very helpfull, because the body diodes of high voltage mosfets are really the most ugly thing to switch... ..also the dedicated Ixys data sheet is not promising for hard switching into a conductive body diode...

...for the design of the filter choke: Take care, the interwinding isolation has to handle high voltage...
 
A high value high voltage inductor can always be divided in smaller series-connected lower voltage inductors. If you have enough room, inductors are never a problem.

Preventing body diode conduction, or proper recovery di/dt (and then dv/dt) control are more interesting problems.
 
Consider realistic parasitic winding capacitances (+ tolerances of L and also of the parasitic capacitances) and drive the series connection with rectangular voltages... You may find the voltages of the nodes between the chokes ringing in unpleasant modes.
I would not generally refuse such series connections, but they are
not naturally free of trouble.

But I agree.
The body diode issue is probably more lethal.
So far I am still convinced that a reasonably high inductive ripple would work here.
 
My idea to attack the diode conduction problem was to add a couple of extra diodes, a very fast shottky, which would not see any high voltage, to prevent any reverse current from flowing into the transistors and turning on the internal diode, plus a HV fast recovery diode to replace the intrinsic one.

ChocoHolic, The actual output current to a resistive load is even lower than 50 mA's, it's only 5 mA's worst case if you consider the electrostatic panel capacitance as a part of the output filter. However I don't fully understand how the output stage seeing an inductive load would prevent the diodes from turning on.
 
Hi Iono,
I should be more accurate.
If you have the inductive load situation this does not avoid currents in the body diode at any time, but it avoids the currents in the body diode at the end of each turn ON phase of the MosFets.
Imagine the steady state idle situation:
Lower MosFet ON and the current is linearly growing through the lower MosFet, means the direction where the body diode of the lower MosFet does not conduct.
Now we turn OFF the lower Mosfet, the choke is still driving the current into the same direction. Consequently the voltage of the half bridge center slopes up until the body diode of the upper MosFet becomes conductive and clamps the sloping. No the voltage across the output inductor has inversed and the current is linearly ramping towards the direction which would be blocked by the body diode of the upper MosFet. But of course we turn ON the upper MosFet before this happens, consequently the current can run through the N-chanel of the upper MosFet. At the time when we turn OFF the upper MosFet the current will already have the direction which is blocked by the body diode and is running in the N-chanel only.
After turning OFF the upper MosFet the voltage will slope down again......

As long as we ensure that this inductive driven current in the switches is higher than the superimposed load current, we will not have the body diodes conductive at the critical moment before the halfbridge voltage slopes.

Very different would be capacitive mode. In capacitive mode the voltage would not slope when you turn OFF the MosFet, but its body diode would conduct the current until the opposite MosFet turns ON. And then the opposite MosFet would have to remove the reverse recovery charge with a heavy current peak.... Such hard switching similar to capacitive mode will also take place if the superimposed load current is larger than the inductive filter ripple.
At lower voltages and high load currents it is less headache to handle the hard switching rather than putting +/-20A filter ripple. But at high voltages the body diodes become more and more catastrophic and your small currents should allow to have a filter current ripple which is larger than the load.

The circuit, which you proposed should also work.
It is traditional and proven in many designs, only drawback is that you add components and parasitic inductances in your PCB.
 
Hi Iono,
I should be more accurate.
If you have the inductive load situation this does not avoid currents in the body diode at any time, but it avoids the currents in the body diode at the end of each turn ON phase of the MosFets.
Imagine the steady state idle situation:
Lower MosFet ON and the current is linearly growing through the lower MosFet, means the direction where the body diode of the lower MosFet does not conduct.
Now we turn OFF the lower Mosfet, the choke is still driving the current into the same direction. Consequently the voltage of the half bridge center slopes up until the body diode of the upper MosFet becomes conductive and clamps the sloping. No the voltage across the output inductor has inversed and the current is linearly ramping towards the direction which would be blocked by the body diode of the upper MosFet. But of course we turn ON the upper MosFet before this happens, consequently the current can run through the N-chanel of the upper MosFet. At the time when we turn OFF the upper MosFet the current will already have the direction which is blocked by the body diode and is running in the N-chanel only.
After turning OFF the upper MosFet the voltage will slope down again......

As long as we ensure that this inductive driven current in the switches is higher than the superimposed load current, we will not have the body diodes conductive at the critical moment before the halfbridge voltage slopes.

Very different would be capacitive mode. In capacitive mode the voltage would not slope when you turn OFF the MosFet, but its body diode would conduct the current until the opposite MosFet turns ON. And then the opposite MosFet would have to remove the reverse recovery charge with a heavy current peak.... Such hard switching similar to capacitive mode will also take place if the superimposed load current is larger than the inductive filter ripple.
At lower voltages and high load currents it is less headache to handle the hard switching rather than putting +/-20A filter ripple. But at high voltages the body diodes become more and more catastrophic and your small currents should allow to have a filter current ripple which is larger than the load.

The circuit, which you proposed should also work.
It is traditional and proven in many designs, only drawback is that you add components and parasitic inductances in your PCB.

Thanks for your explanation. I somehow implied that the output stage was always seeing an inductive load. I will get sure that this happens. However this does not solve the problem as the recovery time is still longer than half the period (I'm thinking in a 400 KHz switching frequency), so it looks it will need the diodes anyways.
 
Well, ...how well your proposed circuit can work at 2.5kV will of course depend on the properties of the fast HV-diode.
Usually 2kV diodes are normally pretty 'ugh'.
May be a series connection of 1kV types is better...
Or may be SiC...

Would a series connection of two diodes withstand twice the voltage?

The SiC devices look really good. I wish one manufacturer made a well-encapsulated 1.7-2.5kV one.
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.