Hi,
You make the mistake of proving that normal amplifier topologies (which work well for resistive loads) don't work for capacitive loads. Which is off course true in most cases. So that's why you need a different design.
Same holds for opamps. Those are really very good for resistive circuits, they're designed around that. Because you need a different amplifier topology for capacitors, they won't work, they're too slow.
I'm not going to give you all the details about my amplifier design. But it basically works like this: You've got a voltage comparator, which gives as output a voltage linearly dependant on the difference between wanted output voltage and the actual output voltage. This voltage comparator steers a CURRENT source which either charges or dischargers the capacitor (depending on if the output voltage is too high or too low)
The only big time-delay in this case is the time it takes to charge the capacitor. If the voltage comparator and current sources can be made sufficiently fast, it will be stable. (in my case there is a time-delay of around 150ns from fault-voltage to correction current; this time-delay determines how much current you can pump into the capacitor per voltage difference)
Normally the voltage comparator immediately steers a voltage source (this is the case for all normal opamps). The problem here is that the voltage-source is probably quite slow, because it involves the heavy output transistors. Oscillations can occur more easily now because there are three delays: voltage comparator (small delay); voltage source starting to deliver current (medium delay); capacitor charging (large delay)
Why would this oscillate more easily?
Well imagine the output voltage is too low. The voltage comparator will quite fast give a high output-voltage. After some time the voltage source will deliver current, and the voltage over the capacitor will start to rise. It will rise faster and faster because the voltage source will give more and more current.
At some point the output voltage is almost OK, but it rises fast. It takes just a little bit of time before the comparator will output that the voltage is OK, and makes the input of the voltage source lower. (at this moment the output is probably already higher dan needed). But now it again takes quite some time to stop the voltage source from delivering current (which at this time is already delivering it's maximum current). So the output voltage will stay rising for quite some time. So there is a large overshoot. At the time the current source has stopped delivering current we're just back at the start, but with a too high output voltage. If the voltage error is now larger then it was at the beginning of the story, the whole thing will oscillate.
If you use a current source instead of a voltage source, then it'll much easier go right; because the output-current will start dropping way before the wanted output voltage is reached.
This makes the output impedance much higher compared to a normal amplifier. Also when you take a resistive output there will be quite a lot of intermodulation-distortion (because of non-linearity of the current sources), but with a capacitive load it works.
Now I must say that I'm not that fond of telling math in a story, the more subtle issues are mostly just forgotten.. I've enlarged the differences in my story a little bit. But I hope you get an idea how it works....
@Bolserst now that's an answer I can do something with.
I had not thought of including the air-weight into this. So this is around 4 times more than the diaphragm weight. It'll halve the resonance-frequency, and indeed with baffle it'll be much more.
Also I had assumed you meant something a little bit different with airload. I thought air-load would be radiating only. But as I said, I don't know very much about acoustics. (always happy to learn 🙂 )
An interesting point is that (without baffle) the airload on the edges of the ESL is much smaller than in the middle. this could cause problems.
I've got to think about this a bit more. Do you know a good source of information about this, with a bit more theory? (so I could include it in my calculations)
High voltage does only remove some stiffness in constant voltage mode, not in constant charge mode. That's why I want constant charge mode 🙂
You make the mistake of proving that normal amplifier topologies (which work well for resistive loads) don't work for capacitive loads. Which is off course true in most cases. So that's why you need a different design.
Same holds for opamps. Those are really very good for resistive circuits, they're designed around that. Because you need a different amplifier topology for capacitors, they won't work, they're too slow.
I'm not going to give you all the details about my amplifier design. But it basically works like this: You've got a voltage comparator, which gives as output a voltage linearly dependant on the difference between wanted output voltage and the actual output voltage. This voltage comparator steers a CURRENT source which either charges or dischargers the capacitor (depending on if the output voltage is too high or too low)
The only big time-delay in this case is the time it takes to charge the capacitor. If the voltage comparator and current sources can be made sufficiently fast, it will be stable. (in my case there is a time-delay of around 150ns from fault-voltage to correction current; this time-delay determines how much current you can pump into the capacitor per voltage difference)
Normally the voltage comparator immediately steers a voltage source (this is the case for all normal opamps). The problem here is that the voltage-source is probably quite slow, because it involves the heavy output transistors. Oscillations can occur more easily now because there are three delays: voltage comparator (small delay); voltage source starting to deliver current (medium delay); capacitor charging (large delay)
Why would this oscillate more easily?
Well imagine the output voltage is too low. The voltage comparator will quite fast give a high output-voltage. After some time the voltage source will deliver current, and the voltage over the capacitor will start to rise. It will rise faster and faster because the voltage source will give more and more current.
At some point the output voltage is almost OK, but it rises fast. It takes just a little bit of time before the comparator will output that the voltage is OK, and makes the input of the voltage source lower. (at this moment the output is probably already higher dan needed). But now it again takes quite some time to stop the voltage source from delivering current (which at this time is already delivering it's maximum current). So the output voltage will stay rising for quite some time. So there is a large overshoot. At the time the current source has stopped delivering current we're just back at the start, but with a too high output voltage. If the voltage error is now larger then it was at the beginning of the story, the whole thing will oscillate.
If you use a current source instead of a voltage source, then it'll much easier go right; because the output-current will start dropping way before the wanted output voltage is reached.
This makes the output impedance much higher compared to a normal amplifier. Also when you take a resistive output there will be quite a lot of intermodulation-distortion (because of non-linearity of the current sources), but with a capacitive load it works.
Now I must say that I'm not that fond of telling math in a story, the more subtle issues are mostly just forgotten.. I've enlarged the differences in my story a little bit. But I hope you get an idea how it works....
@Bolserst now that's an answer I can do something with.
I had not thought of including the air-weight into this. So this is around 4 times more than the diaphragm weight. It'll halve the resonance-frequency, and indeed with baffle it'll be much more.
Also I had assumed you meant something a little bit different with airload. I thought air-load would be radiating only. But as I said, I don't know very much about acoustics. (always happy to learn 🙂 )
An interesting point is that (without baffle) the airload on the edges of the ESL is much smaller than in the middle. this could cause problems.
I've got to think about this a bit more. Do you know a good source of information about this, with a bit more theory? (so I could include it in my calculations)
High voltage does only remove some stiffness in constant voltage mode, not in constant charge mode. That's why I want constant charge mode 🙂
Current/voltage source
Having done both for industrial applications, I do agree with Calvin, it's not trivial to drive capacitive load with voltage feedback amplifier and inductive load with current FB one.
It adds zeros/poles, sometimes unpredictable as well as crossover distortions, due to current flowing at the opposite direction to the voltage applied: transistor or valve can't control it (class B, for instance).
It's a little bit easier to use current feedback for capacitive load but even then the proper current sensor is the pain itself, especially at upper frequency range.
Usual cure is a resistor in series, in order to limit impedance variations/create predictable (easier to fix) pole. There are pretty good app. notes at the Apex web site on this matter.
For "feedbackless" amplifiers load line becomes quite a "circle", one may check what happens to linearity if it is not taken care of.
Said that, striped passive equalizer for wire stators seems to be very elegant solution, limiting unwanted burden on power amplifier.
Alex
Having done both for industrial applications, I do agree with Calvin, it's not trivial to drive capacitive load with voltage feedback amplifier and inductive load with current FB one.
It adds zeros/poles, sometimes unpredictable as well as crossover distortions, due to current flowing at the opposite direction to the voltage applied: transistor or valve can't control it (class B, for instance).
It's a little bit easier to use current feedback for capacitive load but even then the proper current sensor is the pain itself, especially at upper frequency range.
Usual cure is a resistor in series, in order to limit impedance variations/create predictable (easier to fix) pole. There are pretty good app. notes at the Apex web site on this matter.
For "feedbackless" amplifiers load line becomes quite a "circle", one may check what happens to linearity if it is not taken care of.
Said that, striped passive equalizer for wire stators seems to be very elegant solution, limiting unwanted burden on power amplifier.
Alex
Well, having completed the first version of the amplifier some time ago, I can't really say that I had much trouble. It's the first real audio-amplifier I ever designed, and off course it's far from perfect, but there's room for improvements. At the moment I'm happy with it having a (small signal) bandwidth of well over 50kHz, quite stable, and not too much distortion (I don't have the equipment to measure this very thoroughly, I can only say it's below 1% @ 1kHz, cause I can't measure any distortion and this is the measuring precision)
If you say so, I believe you that it's more difficult. To me it looks much harder to design for a resistive load, because the loads vary very much. Here you can just say it's a capacitor of a certain size, and that's all. It's much easier to model and to test. (at least to me...)
Doing this for inductors looks to me much more difficult, because then the load itself has lots of poles, most of them at still important frequencies. Just because inductors aren't very ideal.
I haven't used a very precise current sensor. (actually I just used the input resistance of a transistor, which is far from linear with signal) The precision of the current sensor only has some implications for the high frequencies. Increasing the small-signal bandwidth is probably easier to counter distortion.
At the moment I'm trying to calculate the airflow/resonances around an ESL. I probably won't get anywhere. But I'll give it a try...
If you say so, I believe you that it's more difficult. To me it looks much harder to design for a resistive load, because the loads vary very much. Here you can just say it's a capacitor of a certain size, and that's all. It's much easier to model and to test. (at least to me...)
Doing this for inductors looks to me much more difficult, because then the load itself has lots of poles, most of them at still important frequencies. Just because inductors aren't very ideal.
I haven't used a very precise current sensor. (actually I just used the input resistance of a transistor, which is far from linear with signal) The precision of the current sensor only has some implications for the high frequencies. Increasing the small-signal bandwidth is probably easier to counter distortion.
At the moment I'm trying to calculate the airflow/resonances around an ESL. I probably won't get anywhere. But I'll give it a try...
The airload, or radiation impedance, has both a real and imaginary part.
The imaginary part acts like mass load on the diaphragm.
The real part is what the diaphragm works against to produce sound.
For frequencies below where the wavelength is longer than the diaphragm dimensions:
- the imaginary part (mass load) is essentially constant
- the real part reduces in magnitude with falling frequency, so your SPL will as well without proper equalization
For Acoustic theory I would recommend the book "Acoustics" by Beranek.
You might also find the following AES papers of interest:
- "Radiation Impedance Calculations for a Rectangular Piston", by G. BANK AND J. R. WRIGHT
- "Accurate Model for the Push-Pull Electrostatic Loudspeaker", by PATRICK DE VISSCHERE
Negative stiffness is encountered in both constant voltage mode and constant charge mode.
The difference is that:
- in constant charge mode, the negative stiffness is linear. the force towards the stator increases linearly as the diaphragm moves away from the center resting position; just like a spring. (does not contribute distortion)
- in constant voltage mode the negative stiffness is non-linear. The force toward the stator increases exponentially as the diaphragm moves away from the center resting position. Hence, diaphragm stability is an issue for deflections larger than about 1/3 the D/S spacing. (contributes odd order distortion)
I would highly recommend reading:
“Electrostatic Loudspeakers,” by P. J. Baxandall, Chapter 3 “Loudspeaker and Headphone Handbook, J. Borwick, Ed.
One of the few in depth technical works on ESLs.
To verify the negative stiffness for constant charge mode just charge up your ESL panel and measure response in the near field. Disconnect the HV supply. Monitor the near field response and as charge slowly bleeds off the diaphragm you will see the SPL go down and the frequency of the fundamental resonance peak move up in frequency and Q.
This post shows resonance trends with bias voltage for constant charge mode.
Note that thin felt was applied to the stators to provide damping. Without the felt, the Q was on the order of 5 -10.
Also, a row of silicon dots were applied every 2.5" down the middle of the panel. Without them, the resonance was < 30 Hz.
The SPL level was normalized for all the curves.
http://www.diyaudio.com/forums/planars-exotics/147801-diaphragm-resonance-change-hv-bias.html
After that posting arend-jan pointed me to the Baxandall chapter for which i am very grateful.
Last edited:
I am now reading the book by Baxandal. Great book...
About this constant charge mode: you're right, my fault. Stupid of me, I had calculated the same graphs myself already...
I hope I'll be able to calculate/simulate the response in total. This would be quite a lot of work, I've gotten onto a set of simplified equations, but off course these are still quite complex differential equations. A solution would immediately give the total response of the loudspeaker at any point in space. But I'm afraid this has to ripen a few days in my head before I get any results...
About this constant charge mode: you're right, my fault. Stupid of me, I had calculated the same graphs myself already...
I hope I'll be able to calculate/simulate the response in total. This would be quite a lot of work, I've gotten onto a set of simplified equations, but off course these are still quite complex differential equations. A solution would immediately give the total response of the loudspeaker at any point in space. But I'm afraid this has to ripen a few days in my head before I get any results...
The Finals are the worst sounding ESL's on the market IMO, I'm sure your homemade version if done right will be better ...
Sounds like a quad amplifier from yore .........
Hmm, still people replying to this old thread... Nice.
I must say that at the momement I'm trying to get a simulations of loudspeakers working. I've tried an algorithm using Mathematica, but this is way to slow for usage. Results look hopeful, but I'm still not totally sure it'll give sensible results.
Now I'm trying to program the thing in C/CUDA, it should be quite fast that way. (I need some programming practice anyway for the master thesis I'll have to start soon.)
About this amplifier: PCB's are ordered, but it'll take some time before they get here.
@A. Wayne: I can't compare the sound to other ESLs. But they're very unreliable. The HV-unit has a too weak transformer; the polarizing voltage is a bit too high, so a little bit of dust is a big problem (i'm a student... 😛); the wall-mounting is badly designed (It could have been really good if they took the effort of testing it, but now it takes like ten minutes to get a speaker on it's place); they work in constant voltage mode..
For 100 euros I'm happy, but new they cost around 2000 euro's... It's still a step forward from my KEF104's and DIYspeakers with philips drivers.
@electros: I've thought about it. But I'm a little afraid that this stuff isn't heavy/sturdy enough. (lots of paint could fix that). It does have the big advantage that you can easily make different (insulated) pieces of stator, for different frequencies. I'll keep it in my mind.
First I want to do some simulations, before I'll build a real unit.
I must say that at the momement I'm trying to get a simulations of loudspeakers working. I've tried an algorithm using Mathematica, but this is way to slow for usage. Results look hopeful, but I'm still not totally sure it'll give sensible results.
Now I'm trying to program the thing in C/CUDA, it should be quite fast that way. (I need some programming practice anyway for the master thesis I'll have to start soon.)
About this amplifier: PCB's are ordered, but it'll take some time before they get here.
@A. Wayne: I can't compare the sound to other ESLs. But they're very unreliable. The HV-unit has a too weak transformer; the polarizing voltage is a bit too high, so a little bit of dust is a big problem (i'm a student... 😛); the wall-mounting is badly designed (It could have been really good if they took the effort of testing it, but now it takes like ten minutes to get a speaker on it's place); they work in constant voltage mode..
For 100 euros I'm happy, but new they cost around 2000 euro's... It's still a step forward from my KEF104's and DIYspeakers with philips drivers.
@electros: I've thought about it. But I'm a little afraid that this stuff isn't heavy/sturdy enough. (lots of paint could fix that). It does have the big advantage that you can easily make different (insulated) pieces of stator, for different frequencies. I'll keep it in my mind.
First I want to do some simulations, before I'll build a real unit.
You could take a look at:
Patrick De Visschere
Accurate Model for the Push-Pull Electrostatic Loudspeaker
J. AES, Vol. 43, pp. 563-572, July 1995
Hi,
no I've only built wire stators so far. I was just saying what I'd try to do if I'd want to make such a type of stator
I guess a good choice of paint would be the stuff that is used to repair the defroster on car rear windows. Based on what stuff you can get it might be a good idea to thin it down a little so it won't clog the mesh.
Have fun,
Ken
Here's how I did my polarizing supply. I can recommend it to everyone.
1. Go to a shop that sells PC "case modding" supplies to teenagers
2. Get two sets of "CCFL lamps" (one for each speaker). It's comprised of a 12V connector and switch, a small resonant inverter in a plastic case, a length of silicone-insulated HV wire, and the CCFL lamp proper. It only costs around 5 to 8 EUR for a set!
3. Toss the lamps.
4. The inverter will supply around 1kV AC so you need to put a few stages (maybe 3) of Cockcroft-Walton voltage multiplier ladder behind it.
5. Put a good-old LM317 0-15V (or so) adjustable DC supply in front of it and supply the whole thing with a wall-wart
6. Done!
The resonant inverter runs at about 40kHz unloaded -- depending on how much your speaker leaks HV this will go down to maybe 35kHz under load.
Why is this solution so interesting?
0. Cheap, no transformers needed
1. Electrical safety -- you use a wall wart so no live wires inside your speaker
2. It's an actually useful use of a "case modding" product 😀
3. You can regulate the HV voltage by simply adjusting the LM317 DC supply
4. No hum because of the high frequency
5. You can use way smaller caps in the multiplier ladder, again because of the high frequency. 1nF is probably enough. I bought a bunch of USSR surplus 4n7 4kV film caps for very little money and they are perfect for this.
6. You don't get stung if you accidentally touch a HV wire (though it's probably not very healthy either) 😉
Kenneth
Hi,
beware though that the voltage of those inverters rises to the ignition value of the lamp and drops after ignition. You should use fast diodes in the multiplier. Make shure that there is no audible hiss from the transformer.
Some inverter toplogies need a load connected. Be sure it will be loaded and work when for example the leakage blinker circuit is connected it.
jauu
Calvin
beware though that the voltage of those inverters rises to the ignition value of the lamp and drops after ignition. You should use fast diodes in the multiplier. Make shure that there is no audible hiss from the transformer.
Some inverter toplogies need a load connected. Be sure it will be loaded and work when for example the leakage blinker circuit is connected it.
jauu
Calvin
Well, I think I'm just going to use a simple voltage multiplier directly from the mains. That's quite easy to make. In a few weeks time there is a big radio-amateur-market, where I could buy the capacitors very very cheaply. (1-2 euro's per multiplier)
The high-voltage is feeded through a 10M resistor. It's not that dangerous to touch 1kV through 10MOhm. (only 0.1mA, you can hardly feel it)
It's great to use a simple thing like such a CCFL converter. But you'd still need a 12V supply, and I'm afraid that reliability of 5 euro chinese crap isn't that much better than the original HV-supply. But it's certainly safer.
By choosing the correct output of the multiplier you can still choose your HV in steps of 160V, which ain't that bad. But it'll be dependent on the mains voltage, which isn't too constant (especially where I live).
The high-voltage is feeded through a 10M resistor. It's not that dangerous to touch 1kV through 10MOhm. (only 0.1mA, you can hardly feel it)
It's great to use a simple thing like such a CCFL converter. But you'd still need a 12V supply, and I'm afraid that reliability of 5 euro chinese crap isn't that much better than the original HV-supply. But it's certainly safer.
By choosing the correct output of the multiplier you can still choose your HV in steps of 160V, which ain't that bad. But it'll be dependent on the mains voltage, which isn't too constant (especially where I live).
Not by itself, at least in the units I have. It's just a two-transistor inverter and a ferrite transformer. No ICs, no startup sequences.
There may be a capacitor at the output to make the voltage drop once the lamp starts pulling some current. But it is easily removed.
I looked at a CCFL based design, but decided aginst for the following reasons :
1) The oscillation frequency is high, and I rather avoid HFgetting into my mains if I can.
2) The output voltage is sine wave and not square wave, so there are significant ripples and you need a smoothing cap at the end, rated at say 10kV.
.....Expensive cap.
So I settled for the well known NE555 design at 5kHz. And I can go even lower if I wish. Just bigger caps.
Please don't hesitate to comment. I am new to this game. So I (still) have an open mind.
Patrick
1) The oscillation frequency is high, and I rather avoid HFgetting into my mains if I can.
2) The output voltage is sine wave and not square wave, so there are significant ripples and you need a smoothing cap at the end, rated at say 10kV.
.....Expensive cap.
So I settled for the well known NE555 design at 5kHz. And I can go even lower if I wish. Just bigger caps.
Please don't hesitate to comment. I am new to this game. So I (still) have an open mind.
Patrick
No big cap is needed, because the current draw is so tiny. The ladder capacitors function as smoothing caps, too.
Also, please remember that the series resistor&diaphragm resistivity together with the panel capacitance form a lowpass filter with a very low -3dB frequency.
I'm sure an NE555-based circuit will perform equally well. I just didn't feel like hunting for a suitable transformer that can do 12V->1kV 🙂
Kenneth
At 40kHz you don't even need an HV-capacitor for smoothing. The little capacitance of the diaphragm with the series-resistance has a (when chosen well) a time-constant much larger than the frequency. (The capacitance should be dominant over the whole audible range for constant-charge mode, so if you've got a good design, you don't need extra capacitance)
It's a good frequency especially because you can't hear it. a 5kHz HV-transformer will emit some sound (like old CRT-TV's). That's something i really wouldn't recommend!
10kV capacitors are not really expensive, they're hard to get. But radio-amateurs sometimes use them for heavily resonant antenna's. 5 euros is a normal price for ~1nF.
Don't care about having some HF on your mains. Ever put the mains on an oscilloscope? It does have some similarities with a sine, but there is a LOT of HF on it. Nice thing is: It doesn't really matter. (unless it gets REALLY bad. But this won't be a problem)
@Calvin: The voltage rises to ignition value, and falls when there's a CCFL connected, because the CCFL's resistance drops a lot when it's ignited. But when there's no CCFL connected the voltage stays high. (normally)
edit: @kavermei: You can also use a flyback-transformer. look it up on the internet. You can get high votages easily with those things, without having very high winding ratios.
(or just get a junkbox full of transformers like me 😛)
It's a good frequency especially because you can't hear it. a 5kHz HV-transformer will emit some sound (like old CRT-TV's). That's something i really wouldn't recommend!
10kV capacitors are not really expensive, they're hard to get. But radio-amateurs sometimes use them for heavily resonant antenna's. 5 euros is a normal price for ~1nF.
Don't care about having some HF on your mains. Ever put the mains on an oscilloscope? It does have some similarities with a sine, but there is a LOT of HF on it. Nice thing is: It doesn't really matter. (unless it gets REALLY bad. But this won't be a problem)
@Calvin: The voltage rises to ignition value, and falls when there's a CCFL connected, because the CCFL's resistance drops a lot when it's ignited. But when there's no CCFL connected the voltage stays high. (normally)
edit: @kavermei: You can also use a flyback-transformer. look it up on the internet. You can get high votages easily with those things, without having very high winding ratios.
(or just get a junkbox full of transformers like me 😛)
Last edited:
The circuit I am using uses a 230V / 6V 1VA EI that costs next to nothing; then setp up with voltage multiplier ladder.
Patrick
Patrick
This reminds me of the fun I had with a TV flyback when I was a bit younger... I still remember the blue glow around my fingers and the smell of ozone 🙄
BTW, where and when is the HAM fair?
Kenneth
This fair: 50e VERON Dag voor de Radio Amateur (DvdRA) 2010
(Hmm, it's in November, thought it was earlier)
If you're good at searching for bargains, there's much to be had. Lots of (old) electronics components (vacuum tubes, HV-caps etc etc), lots of measuring equipment (not very cheap, but there are always some good deals...), and lots of radio-stuff, (antenna's transmitters, communication-receivers etc.).
But don't expect much directly audio-related things, it's a radio-electronics-fair. So only if you build electronics yourself, then it's worth the bother.
(Hmm, it's in November, thought it was earlier)
If you're good at searching for bargains, there's much to be had. Lots of (old) electronics components (vacuum tubes, HV-caps etc etc), lots of measuring equipment (not very cheap, but there are always some good deals...), and lots of radio-stuff, (antenna's transmitters, communication-receivers etc.).
But don't expect much directly audio-related things, it's a radio-electronics-fair. So only if you build electronics yourself, then it's worth the bother.
- Status
- Not open for further replies.