I'm building a pair of electrostatic loudspeakers following some easy recipes I've found on the net. While they are rather easy to build, the big problem seems to be driving them, as they aren't exactly a nice load and they require a step-up transformer with it's distortion and bandwidth problems.
Since I've been helping friends in building high-voltage power supplies for some time, it seems pretty easy to me to build a "controlled" high voltage power supply to power the ESL's, that's it, a class D amp. I have found no information on the net about this kind of amp so it's either a very crank idea or I'm the first one to seriously try this.
My approach would be:
Triangle wave generator -> comparator -> mosfet -> flyback transformer
Audio input -> summing amplifier (bias) -> to comparator
Switching frequency would be around 200KHz. I believe that this setup could sound OK with no feedback at all, but this has to be seen.
Any input?
Since I've been helping friends in building high-voltage power supplies for some time, it seems pretty easy to me to build a "controlled" high voltage power supply to power the ESL's, that's it, a class D amp. I have found no information on the net about this kind of amp so it's either a very crank idea or I'm the first one to seriously try this.
My approach would be:
Triangle wave generator -> comparator -> mosfet -> flyback transformer
Audio input -> summing amplifier (bias) -> to comparator
Switching frequency would be around 200KHz. I believe that this setup could sound OK with no feedback at all, but this has to be seen.
Any input?
Hi,
I don´t think the idea is crank and You´re definitely not the first to think in that direction. You´d be probabely the first if You managed to build a working -and of course legally(!) working- item
On the Pro-side there are:
- the capability of switching amps to control even very complex loads like ESLs with ease.
- their inherent high efficiency.
On the Con-side there are:
- finding fast enough switching power transistors or circuits that are capable of at minimum 2kV, preferable 4-5kV, with large ESLs even up to 10kV supply voltage.
- EMV-rules and - compatibility is already not that easy to accomplish with typical classD, let alone highvoltage-classD.
- distortion values will probabely be higher than those of a well executed ESL panel driven by a first class audio-transformer. Why spoil the performance? Sonically every class-D I listened to performed rather uninspiring and dry. A good tube amp seems still be to be the best sounding match, wether in Direct Drive or coupled via audio transformer.
Assumptions You made, which imho don´t hold up in practise:
- ESLs are easy to build. No they aren´t! It´s easy to build a sound generating device with an electrostatic motor. A decent ESL is something very different. If it were not, then there would be many more manufacturers who would try this superior loudspeaker principle. With ESLs the devil is in the details.
- audio transformers restrict in bandwidth and distortion. That is true, but a well designed audio tranny allows for such good values that most direct drive amlifiers don´t reach better results either. If You look at common directdrive designs there is hardly one that reaches a full power bandwidth of just 20kHz!
- Apart from some enigmatic devices fabricated from unobtainium, there are hardly any output devices on the market that might be usable as power output stage. IXYS specialized a bit on highvoltage type transistors up to 4kV (IGBTs), but I doubt You get those to switch cleanly at 200kHz and 4kV. With the upcoming of the SIC-technology new power-JFETs which look quite promising may make it to the market. But to my knowledge there are only two companies that offer samples at the moment at considerable high costs. Both restricted to 1200V Uds at the time. One offers enhancement, the other company depletion mode N-channel types. Stacked devices switches with up to 6.5kV have already proven their functionality using the latter devices. Especially those dpletion mode JFETs look very promising for a common classA/AB directdrive design with stacked devices. There might be some transmitting tubes, but at considerable cost too.
jauu
Calvin
I don´t think the idea is crank and You´re definitely not the first to think in that direction. You´d be probabely the first if You managed to build a working -and of course legally(!) working- item
On the Pro-side there are:
- the capability of switching amps to control even very complex loads like ESLs with ease.
- their inherent high efficiency.
On the Con-side there are:
- finding fast enough switching power transistors or circuits that are capable of at minimum 2kV, preferable 4-5kV, with large ESLs even up to 10kV supply voltage.
- EMV-rules and - compatibility is already not that easy to accomplish with typical classD, let alone highvoltage-classD.
- distortion values will probabely be higher than those of a well executed ESL panel driven by a first class audio-transformer. Why spoil the performance? Sonically every class-D I listened to performed rather uninspiring and dry. A good tube amp seems still be to be the best sounding match, wether in Direct Drive or coupled via audio transformer.
Assumptions You made, which imho don´t hold up in practise:
- ESLs are easy to build. No they aren´t! It´s easy to build a sound generating device with an electrostatic motor. A decent ESL is something very different. If it were not, then there would be many more manufacturers who would try this superior loudspeaker principle. With ESLs the devil is in the details.
- audio transformers restrict in bandwidth and distortion. That is true, but a well designed audio tranny allows for such good values that most direct drive amlifiers don´t reach better results either. If You look at common directdrive designs there is hardly one that reaches a full power bandwidth of just 20kHz!
- Apart from some enigmatic devices fabricated from unobtainium, there are hardly any output devices on the market that might be usable as power output stage. IXYS specialized a bit on highvoltage type transistors up to 4kV (IGBTs), but I doubt You get those to switch cleanly at 200kHz and 4kV. With the upcoming of the SIC-technology new power-JFETs which look quite promising may make it to the market. But to my knowledge there are only two companies that offer samples at the moment at considerable high costs. Both restricted to 1200V Uds at the time. One offers enhancement, the other company depletion mode N-channel types. Stacked devices switches with up to 6.5kV have already proven their functionality using the latter devices. Especially those dpletion mode JFETs look very promising for a common classA/AB directdrive design with stacked devices. There might be some transmitting tubes, but at considerable cost too.
jauu
Calvin
I believe that in the flyback topology the power transistor will not have to withstand the full output voltage as the load is connected to the secondary of the transformer.
The idea of using high voltage tubes looks really good. I will look for class D tube designs.
About whether it's easy to build a ESL loudspeaker or not, I've still to decide if I'm doing an audio project or a science projects. Probably it will start like a science project and turn into an audio project after the first working results.
My initial idea was to use a TV flyback output transformer, but I'm not sure if it will last long driving an ESL.
The idea of using high voltage tubes looks really good. I will look for class D tube designs.
About whether it's easy to build a ESL loudspeaker or not, I've still to decide if I'm doing an audio project or a science projects. Probably it will start like a science project and turn into an audio project after the first working results.
My initial idea was to use a TV flyback output transformer, but I'm not sure if it will last long driving an ESL.
For half or full bridge class D I think there are useable MOSFET and diodes up to 1500V. A 1500V full bridge could drive the panels directly, or drive a step-up transformer with a low turns ratio, which should perform much better in terms of frequency response, distortion and parasitic capacitance than conventional audio transformers. But getting a switching stage right at 200V is already tricky, at 450V (the highest I have done) it's hard and frightening every time you get a set of exploded output transistors, and at 1500V add "risk of death" to the equation...
Flyback is a different matter, I would not recommend it for audio, transfer characteristic is not linear, output voltage has a "1/(1-duty_cycle)" term in it that makes it highly non-linear.
Anyway, audio power transformers with high step up ratios are terrible, most of the difficult load associated with electrostatic speakers actually comes from the capacitive and resonant nature of the transformer. Class D amplifiers tend to behave much better. Nowadays many designs have load independent frequency response (the one in which I'm working changes less than 0.1 dB from no load to full load) and low THD.
Most tube amplifiers with output transformer will easily show several +/-3dB peaks and dips across the audio band depending on load, thanks to their impressive unity damping factor
Flyback is a different matter, I would not recommend it for audio, transfer characteristic is not linear, output voltage has a "1/(1-duty_cycle)" term in it that makes it highly non-linear.
Anyway, audio power transformers with high step up ratios are terrible, most of the difficult load associated with electrostatic speakers actually comes from the capacitive and resonant nature of the transformer. Class D amplifiers tend to behave much better. Nowadays many designs have load independent frequency response (the one in which I'm working changes less than 0.1 dB from no load to full load) and low THD.
Most tube amplifiers with output transformer will easily show several +/-3dB peaks and dips across the audio band depending on load, thanks to their impressive
unity damping factor
Last edited:
i guess that a ignition coil from a car would be able to deliver more power than one from a TV-set, but i also think both would be underrated in terms of power.
I will follow this thread, it would be an interesting amplifier, especially if you would combine class-D with a tube, new and old technology integrated in one amplifer
I will follow this thread, it would be an interesting amplifier, especially if you would combine class-D with a tube, new and old technology integrated in one amplifer
A phase shifted full bridge driving a step-up or step-down transformer, and then an output filter, can make a class D amplifier with reasonably good open loop linearity. Doing it that way, the transformer is not subject to audio frequencies. Peavey (and probably others) patented it a long time ago. But the problem is rectification: Polarity of output winding has to be flipped on every cycle too, which requires active devices on the secondary side to achieve "four quadrant" I/V operation, not a good thing at high output voltages.
Same happens with flyback.
An audio amplifier must be able to both source and sink current regardless of the polarity of output voltage, particularly with a capacitive load like ESL. A flyback can only work in one quadrant. Half bridges and full bridges can be easily operated in 4 quadrant mode (conventional class D).
Same happens with flyback.
An audio amplifier must be able to both source and sink current regardless of the polarity of output voltage, particularly with a capacitive load like ESL. A flyback can only work in one quadrant. Half bridges and full bridges can be easily operated in 4 quadrant mode (conventional class D).
Last edited:
So it seems that the biggest problem is rectification, as i'm thinking about using a triode (Something like a PD500, I have some of these lying around) to sink current, this would make the amplifier extremely inefficient and will not be an elegant solution but it will allow some experimentation with the class-D part working on a single quadrant and doing the hard "voltage lifting" part.
The tube seems a nice part to experiment as it can withstand enormous voltages, the only concern being x-ray emission.
Any suggestions for rectification part?
The tube seems a nice part to experiment as it can withstand enormous voltages, the only concern being x-ray emission.
Any suggestions for rectification part?
You could do a half bridge with triodes or pentodes as switches, tube or solid state diodes and solid state grid driving, operating at several kV
I wonder if tubes can switch fast enough with proper grid drive... Less than 250ns total switching delay would be good, 500ns would probably be the maximum acceptable for 20khz class D.
I wonder if tubes can switch fast enough with proper grid drive... Less than 250ns total switching delay would be good, 500ns would probably be the maximum acceptable for 20khz class D.
You could do a half bridge with triodes or pentodes as switches, tube or solid state diodes and solid state grid driving, operating at several kV
I wonder if tubes can switch fast enough with proper grid drive... Less than 250ns total switching delay would be good, 500ns would probably be the maximum acceptable for 20khz class D.
I will give a try to the switching times soon, I've somehow believed that the high-frequency capability of tubes meant there would be no problem with it. Now I'm trying to build a simple model for the loudspeaker to have an idea of the drive requirements.
About the use of tube rectifiers, aren't they too nonlinear? Solid state rectifiers are nonlinear too but the voltage drop in this application is negligible compared to the output voltage.
It turns out that the PD500 is a shunt regulator, so maybe using it for that purpose is not not that crazy.
Hi,
the model for the speaker could be a lowloss cap. The capacitance depens on stator-area and stator-distance in first place (C= 8,854*10e-12*A/d || C in F, A in m², d in m). Some of the commercial designs settle around 1nF. What makes this load difficult is its impedance which is complex in nature and of falling value with rising frequency. So with rising frequency we need increasingly more power (more current), which is not what amplifiers like. Since there is no series resistance as in a voicecoil of a dynamic speaker efficiency can be very high (up to 30% have been reported), but falls with increasing frequency. Driving capacitive load with sine waves You need Ipeak=Cges*pi*f*Vp-p || Cges= Cesl+Ccable+Camp
V-pp= 4*d/s [kV] ||d/s diaphragm-to-stator-distance = 0.5*stator-to-stator-distance.
Signal voltages quickly rise with increasing stator-stator-distance.
For a hybrid-ESL-panel working from approximately 300Hz on a d/s of 1mm is sufficient. For lower frequencies larger d/s-values are needed. This small d/s value already requires up to 4kVp-p of signal voltage.
Calculating with values of 1nF and an bandwidth of 20kHz, Ipeak equals 250mA. So to keep power demands smaller you can either:
- fudge with lower bandwidth (the power distribution in a music signal is falling with rising frequency)
- decrease capacitance (by electrical segmentation)
- driving the panel at lower levels and below its requirements, thereby accepting clipping every now and then.
With 1500V-parts (which might be used to maybe 1200V reliably) as output transistors even a bridge connected amplifier would be hardly sufficient to drive a panel with the above suggested parameters. Those transistors are rather good for ESL-headphones which require less signal voltage and currents, but not for a decent loudspeaker-panel.
You need either to stack them or use other more capable devices.
jauu
Calvin
the model for the speaker could be a lowloss cap. The capacitance depens on stator-area and stator-distance in first place (C= 8,854*10e-12*A/d || C in F, A in m², d in m). Some of the commercial designs settle around 1nF. What makes this load difficult is its impedance which is complex in nature and of falling value with rising frequency. So with rising frequency we need increasingly more power (more current), which is not what amplifiers like. Since there is no series resistance as in a voicecoil of a dynamic speaker efficiency can be very high (up to 30% have been reported), but falls with increasing frequency. Driving capacitive load with sine waves You need Ipeak=Cges*pi*f*Vp-p || Cges= Cesl+Ccable+Camp
V-pp= 4*d/s [kV] ||d/s diaphragm-to-stator-distance = 0.5*stator-to-stator-distance.
Signal voltages quickly rise with increasing stator-stator-distance.
For a hybrid-ESL-panel working from approximately 300Hz on a d/s of 1mm is sufficient. For lower frequencies larger d/s-values are needed. This small d/s value already requires up to 4kVp-p of signal voltage.
Calculating with values of 1nF and an bandwidth of 20kHz, Ipeak equals 250mA. So to keep power demands smaller you can either:
- fudge with lower bandwidth (the power distribution in a music signal is falling with rising frequency)
- decrease capacitance (by electrical segmentation)
- driving the panel at lower levels and below its requirements, thereby accepting clipping every now and then.
With 1500V-parts (which might be used to maybe 1200V reliably) as output transistors even a bridge connected amplifier would be hardly sufficient to drive a panel with the above suggested parameters. Those transistors are rather good for ESL-headphones which require less signal voltage and currents, but not for a decent loudspeaker-panel.
You need either to stack them or use other more capable devices.
jauu
Calvin
To take profit of the high efficiency intrinsic to the design one must have an amplifier capable of pumping back current from the loudspeaker to the reservoir capacitors, which I believe to be almost impossible.
About the model of the loudspeaker I'm trying to build, my idea is to include energy radiated as sound. The starting point was ultra-simple, I just calculated the electric fields inside the loudspeaker by making simple assumptions, used the energy density formula u=1/2(epsilon0)*E^2 and the enclosed volume to get the potential energy and differentiated with respect to the diaphragm position to get the force, which I related to the velocity by using the air's acoustic impedance.
My idea is that this model could be used to get rough estimates of sensitivity, frequency response and load impedance as well as a theoretical limit to the sensitivity given by the air dielectric strength. So far this is not working, I believe due to a simple calculation mistake. I will spend some more time around it, as I like the energy density approach to solve that kind of problems because it lends itself well to more advanced treatments like normal mode decomposition for the diaphragm or finite element simulation. I hope to get somewhere soon to get back to the electronics again.
My idea is that this model could be used to get rough estimates of sensitivity, frequency response and load impedance as well as a theoretical limit to the sensitivity given by the air dielectric strength.
For a rough estimate of the sensitivity you could use what could be called 'Peter Walker's equation' which predicts the sound pressure in the far field give the signal drive current:
P = i * Vpol / (2 * pi * r * c * d)
with:
i = signal drive current
Vpol = diaphragm polarizing voltage
c = sound velocity
r = distance to speaker
d = diaphragm to stator spacing
Frequency response can't be calculated from macro expressions like energy density, you have to account for real world behavior like resonance of the diaphragm. But given a light enough diaphragm and ignoring the main resonance you should expect a +6dB/octave slope with rising frequency for a voltage driven panel and a flat response for a current driven panel. Assuming an unsegmented flat panel that is. As a result of the increasing directivity with frequency you need a lot less power in the high frequencies.
The input impedance can be seen as a pure capacitor as a first order approximation.
I'm interested to see where this goes, I've always had a fetish for a direct drive amplifier for an ESL but only met people who said it can not be done (well).
For a rough estimate of the sensitivity you could use what could be called 'Peter Walker's equation' which predicts the sound pressure in the far field give the signal drive current:
P = i * Vpol / (2 * pi * r * c * d)
with:
i = signal drive current
Vpol = diaphragm polarizing voltage
c = sound velocity
r = distance to speaker
d = diaphragm to stator spacing
Frequency response can't be calculated from macro expressions like energy density, you have to account for real world behavior like resonance of the diaphragm. But given a light enough diaphragm and ignoring the main resonance you should expect a +6dB/octave slope with rising frequency for a voltage driven panel and a flat response for a current driven panel. Assuming an unsegmented flat panel that is. As a result of the increasing directivity with frequency you need a lot less power in the high frequencies.
The input impedance can be seen as a pure capacitor as a first order approximation.
I'm interested to see where this goes, I've always had a fetish for a direct drive amplifier for an ESL but only met people who said it can not be done (well).
Yeah, but this happens in conventional class-D circuits where it is not possible to raise the voltage past the supply rails. I'm in love with the idea of having an analog low voltage front-end with a high voltage output. This adds up to the nonlinearity problem. My experiences with power IGBT's make me crave for a low-voltage switching stage.
Would a system with a step-up transformer still be able to pump back current to the source if rectification is done after the secondary?
BTW, Calvin, where was the 70% of the energy going in this "super-efficient" design? Was it a transformer-loudspeaker combo?
My equations show there is an elastic restoring force and an electric restoring force. A model that doesn't take into account diaphragm resonances is assuming that the electric restoring force is far bigger than the elastic force. This is like it should be because the elastic force is frequency-dependant since different normal modes of vibration will have different elastic constants. I would like to work a bit more on it to see where it gets.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.