Another direct drive thread

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I see that some people have been working on DD here. I've built a few myself as well, all using mosfets.
My goal has always been building something that is practical in everyday life, an affordable reliable amp capable of driving a fullrange ESL to reasonable SPL levels and sounding better than a good trannie. Not some 2000W dissipating beast or some compromised amp with severe limitations in output voltage/current.
I'd like to throw in some of my own ideas and experiences, just to tease the mind and maybe trigger some discussion that will raise a few good ideas :)

What do we need:
Voltage: a 1:100 stepup transformer connected to a 100W@8ohm amp will give 8000Vpp and this is definately no overkill.
Using a conventional stepup transformer I measured actual currents flowing in the stator while playing different kinds of music and found amazing high (short) peaks up to several hundreds mA. RMS of course is much less. This is a bit more than might be expected from U/Z but that might be explained by the reactive nature of an ESL (it can store energy). To sum it up, 25 mA for each kV will cover most needs and is also what I found reasonable when testing different class A direct drive designs.
So 8000V and 200mA (peak to peak) is the goal. An efficient bridge amp running from 4000V rail will provide enough voltage.

When designing something like this it's a good thing to take safety/reliability into consideration from the start. Demand for a 0 V DC potential at the outputs is obvious as we can't rely on stator isolation for our lifes. Safety in case of failure is another point, the amp should better not link 4000V to your preamp controls when something does blow up. Fuses won't do, anyone who experienced a fuse blow trying to break 4 kV knows that this is a very bad idea :cannotbe: Also failure of any kind must not result in parts on the outside of the casing becoming live. Which implies that heatsinks have to be inside the enclosure. Power dissipation must therefore stay within reasonable limits (~100W/ch). Which is also needed to keep running costs reasonable.

What to use?
- MOSFETS seem the way to go. No second breakdown so very reliable in HV. Cheap, efficient, available up to 1500V @ several amps. To handle 4000V they can be stacked, using resistor dividers. The resistors in combination with internal capacitances however limit achievable bandwidth which put a practical limit on the amount of devices that can be stacked. It works fine up to 4 or 5 fets in my experience. Linearity of mosfets is poor, especially at high voltages due to internal capacitances that are voltage dependent, but what can we do..
- Tubes: Linearity is probably better than mosfets. There are even a few that will handle 4kV. But efficiency will be very poor indeed due to their filament requirements and severely limited output swing which will require a power supply voltage much higher than 4kV, increasing dissipation even more. There's also the problem of filament-to-cathode isolation in push-pull designs.
- Bipolair HV transistors suffer from serious SOA restrictions that limit their practical use to a few hundred volts. Also hv-types have low beta and poor linearity. Low beta means stacking is troublesome, darlingtons are mandatory doubling component count in the output stage.
- Large HV-IBGT's tend to be overrated for what we need (200 A is a bit overkill) and for audio these massive devices are too slow due to the high internal capacitances.

Possible amplifier concepts:

SE is equivalent to class A so this can be ruled out right away. Even with constant current source loading you're looking at 800-1600W dissipation for one channel.

SRPP (such as Acoustat used) will reduce dissipation by a factor 2 at best, probably much less when the load isn't purely resistive. Not very helpfull.

Class D: It will be very difficult to construct an output stage capable of switching 4000V at 200khz or more. Think of parasatic capacitances that have to be charged/discharged requiring massive currents. Efficiency might end up pretty low. Maybe use LC resonance to increase output voltage? There is also the issue of EMC. Generating high speed square waves at several kV may not be the best starting point for an amp that will not will interfere with the rest of the audio equipment (and block out radio reception in the entire neighbourhood). Massive filtering is needed and all filter components will have to be rated for the full output voltage and current. Might be possible but not very practical.

Class AB seems the way to go. But AB requires a complementary output stage and none of the available active devices come in two polarities, at least if high voltage rating is required. Still it can be done using a large stack of complementary fets for lower voltage. The best pair I found so far is good for 400V (IRFU310/9310). So stacking 10 of each building a 20 fet stack for 4 kV? Nope, bandwidth will be too low. Even in spice (which tends to be optimistic with such designs) I wasn't able to achieve closed loop bandwidth of more than a few khz, not to mention slewing problems and poor symmetry. Bandwidth can be increased by decreasing the voltage divider resistors lowering RC times, but this results in unacceptable large idle currents and dissipation. Additional problem with large stacks is poor reliability: During fast transients voltage division between the fets is determined bij internal capacitances rather than the resistor values. These capacitances have huge tolerances and are also voltage dependent.

One way around the N/P problem is a SE stage followed by a source follower. This will be able to deliver more current than idle setting. The concept works, I built it, running @ 4kV and 10mA idle current, and it's very reliable. But it is also extremely non-linear. No practical amount of feedback can cure that without compromising stability too much and I consider it to be sounding much worse than a good transformer. Unless huge dc-blocking capacitors are added it also won't meet safety demands as it places 2000V DC on the esl plates.

True push pull then.. Using optocouplers it is possible to transfer a control voltage to the top fet of an output stage using 2 n-channel fets, creating true pp operation. It allows for symmetrical power supply with 0V dc out. I actually built this but failed to get it to work properly. Problem is how to gain accurate control over the upper fet while it is happily swinging several kV's up and down on the output rail. The full voltage swing present over the optocoupler togehter with parasatic capacitances in the optical path make it impossible to control the fet in an accurate way.
Maybe it can be done using a self-made optocoupler device using fiberoptics. But still there are high noise levels and non-linearities with optocouplers.
The same can be achieved using transformers but problems will be similair. Possible way around it is to use fully differential signal transfer but I doubt if enough CMMR can be achieved to obtain any accurate control over the upper fet. Still, might be worth a try.

Class h/g: is an extension of AB: can't be built properly using one polarity output devices. The concept might still be usefull for reducing dissipation in a class A design, but I think complexity will be too high (and thus reliablity too low).

Lately I've been looking at the concept of pp amp running at say 1000V feeding an esl-transformer with low stepup, possibly within the feedback loop. But I learned that a lower stepup will not make things easier on the trannie as the problems are caused by output voltage and bandwidth demands rather than by the stepup ratio.

So after a few years of experimenting and thinking I am back at the beginning. I start to believe that building a practical DD amp that has more advantages than disadvantages is not possible. After all, there are none on the market and there must be some good reason for that ;)

But hopefully someone can prove me wrong? :D

Regards,

Martin
 
maudio said:
Lately I've been looking at the concept of pp amp running at say 1000V feeding an esl-transformer with low stepup, possibly within the feedback loop. But I learned that a lower stepup will not make things easier on the trannie as the problems are caused by output voltage and bandwidth demands rather than by the stepup ratio.

I think this is the best approach. I think the problems for the transformer are due to the large amount of wire that has to be on the secondary of a 1:100 XFMR. A lower turns ratio means a lot less wire, a lot less leakage inductance (I think) and less capacitive coupling between the primary and secondary.

I think a wide bandwidth, linear 1:10 XFMR is easier to make than a 1:100 XFMR.

I_F
 
Re: Re: Another direct drive thread

I_Forgot said:


I think this is the best approach. I think the problems for the transformer are due to the large amount of wire that has to be on the secondary of a 1:100 XFMR. A lower turns ratio means a lot less wire, a lot less leakage inductance (I think) and less capacitive coupling between the primary and secondary.

I think a wide bandwidth, linear 1:10 XFMR is easier to make than a 1:100 XFMR.

I_F

Hi,
I would not agree. You would have to use even more wire(in length) for a 1:10 step-up , than 1:100. For example if you have 50 prim. and 5000 secondaries for 1:100 step up , for the same core and 1:10 step-up , you would need 500 prim. and still 5000 secondaries.
Otherwise , the trafo would saturate much quicker.

Regards,
Lukas.
 
Re: Re: Another direct drive thread

I think the problems for the transformer are due to the large amount of wire that has to be on the secondary of a 1:100 XFMR. A lower turns ratio means a lot less wire, a lot less leakage inductance (I think) and less capacitive coupling between the primary and secondary.

I think a wide bandwidth, linear 1:10 XFMR is easier to make than a 1:100 XFMR.

If only that would be true, but it isn't. Lowering stepup will not allow you to reduce the # of turns. In fact, you'll need more turns as the primairy has to have more turns to handle the higher input voltage.

Look at it this way: Given a certain core we are stuck with a certain limit on the total flux it can handle, dependent on core size and u.

Combined with the lowest frequency demand this will determine the volt/turn ratio we can squeeze out of this transformer.

The v/turn ratio together with required output voltage then determines the # of secundairy turns we need.

Low or high stepup will only change the # of primairy turns, lower stepup will mean more turns which actually is making things worse (more capacitances).

The bandwidth at the high end is set by leakage inductance and the (to the primairy reflected) internal + load capacitances. Reducing stepup will decrease the capacitance seen at the primairy but leakage will also increases as the # of prim turns will be higher. The bottom line is you'll end up at best with exactly the same bandwidth restrictions again, probably even worse because of the higher capacitances.
 
So after a few years of experimenting and thinking I am back at the beginning. I start to believe that building a practical DD amp that has more advantages than disadvantages is not possible. After all, there are none on the market and there must be some good reason for that

Take it from the guy who built that impractical 2000W dissipating beast you're probably referring to, I think you're right. DD is largely impractical. My efforts were a personal Everest, if you will. Plus a test bed for ESL development, or so I told my wife. Having climbed the summit, I don't need to do it again. And yes, I did see bodies along the way...

The one exception, and it's not really even DD: If you prefer tube amps, as do all perceptive audiophiles :D , you could use a low turns-ratio transformer combining both output transformer duty and step-up duty, all in one device. And yes, one well designed 1:3 (or so) transformer should have better performance than the combination of one regular tube output transformer and one 1:100 step-up transformer. Now that makes some practical sense to me.
 
And yes, one well designed 1:3 (or so) transformer should have better performance than the combination of one regular tube output transformer and one 1:100 step-up transformer.

I agree on this one, if you insist on tube sound it is a logical step to combine 2 transformers into 1.
But IMHO this has nothing more to do with direct drive than using a 1:100 stepup after a solid state amp. As I explained above, from a transformer performance point-of-view there is no significant difference between 1:3 or 1:100 stepup. Apart from the tube-vs-silicon thing (let's not restart that here ;) ) I don't see any advantage in one over the other.

Or it must be the fact that for driving esl's (with or without transformer) you'll need all the power you can get, which might give solid state some advantage?

Anyway, I guess we all keep on waiting for some manufacturer to start producing that magical hv-p-channel fet ...

(or for the discovery of positive electrons, if you're into the tube thing)

:D
 
Hi ,
I think the best solution is to use an amplifier running from a higher voltage and use a lower turns ratio transformer . Say for instance you have a push pull valve (or high voltage mosfet amp) . Try using a valve output transformer with ultra linear taps , drive from the UL taps and take the output from the anode connections . This seems to work very well . For smaller treble panels toroidal mains transformers can be used wired as autoformers , I'm using a 110+110:55+55 running from a small valve amp wired like this

output->110-55-0-55-110<-output

inputs on 110-55v 'taps' of the transformer

cheers

316a
 
Simu probs

Hi,

I found this schematic on the net, which looked good to me on first glance. But I couldn´t get a simu running well with this circuit.
An externally hosted image should be here but it was not working when we last tested it.

First of all I changed The LM318 and the TLC for OPA604 and second I changed the 15V-Diode D1 to a voltage source since with the models in the CAD-prog no Diode would give 15V.
Now I´m a bit confused about the positive bias voltage coming from VR1. This positive offset goes together with the audio signal to the negative input of U4. So Q1 that needs a positive bias too will get a negative bias.
Any Ideas?

jauu
Calvin
 
Ex-Moderator
Joined 2003
The big advantage of the low step-up ratio output transformer approach is that it avoids the problem of complementary devices and allows you to have true push-pull with identical devices (whether they be valves, MOSFETs, or whatever). The second advantage of using an output transformer is that it allows you to swing to twice rail voltage, and that increases your efficiency.

The big problem with transformers is octaves of bandwidth. But that's also true of any loudspeaker. An ESL panel that can produce bass requires a large area, large gap and therefore large voltage swing. A panel to produce treble can get away with a small gap (lower voltage) and small area (to reduce the capacitance). Since you probably have a bass panel and a treble panel, do you really need a single full-range amplifier? If not, the transformer problems for a dedicated bass amplifier and a dedicated treble amplifier are much reduced.
 
Calvin, this is the McKean amp that keeps turning up on the internet now and then. It is a single ended amp with either a resistive load or a or constant current source (as in the one you posted). It's limited to 1000V out and if I am not mistaken, around 40 mA. However I love the way he constructed the CCS. The offset voltage is there to regulate the dc-output at 1/2 Vcc.
The output voltage is 8 times (= 18 dB!) less than a 1:100 trannie on a decent amp. Fine for a headphone, just won't do on a large panel. I built several amps running from 1 kV rails and they are just not capable of driving larger panels at any realistic levels.

One of the better ones I built is attached (posted it earlier in another thread), delivers close to 2kV tt, max current is around 12 mA with shown component values but when you increase idle current to around 50 mA (R4,10=22ohm, R3,12 = 56 ohm) and provide HUGE heatsinking it will drive a large panel, just not very loud. This one runs in simulation as well as in the real world ;) Don't forget to add gate-protection zeners, they are missing in the schematic.
CCS is much more simple but works adequate. With HV-designs the keep-it-simple principle is very valuable... By using leds that will zener reliable at much lower currents I overcame the zenerproblem Calvin mentioned.


The philips design is very interesting, in fact philips developped a full working esl-system back then. Unfortunately it never made it to the outside world. Don't know why they decided to scrap it but that's the way things often go it seems...

From what I heard about it, it worked actually very good. I believe they used optically driven HV-transistors (or diodes, for that matter) that were developped/manufactured specially for this project. So little chance of reproducing it :( Still, might be an interesting project to drill holes in some HV-diodes and see what happens :D

I believe the engineer running the project was a Mr. Streng who published several other interesting papers about esl's.
 

Attachments

  • hvamp.pdf
    28.5 KB · Views: 249
Hi ec8010 cat,

EC8010 said:
The big advantage of the low step-up ratio output transformer approach is that it avoids the problem of complementary devices and allows you to have true push-pull with identical devices (whether they be valves, MOSFETs, or whatever). The second advantage of using an output transformer is that it allows you to swing to twice rail voltage, and that increases your efficiency.
Fully agree on that, and don't forget the additional isolation (read: safety) the transformer provides and which is very welcome is such designs.
Actually, I have been experimenting with this idea a while ago, I wound a 1:6 transformer and built an amp around it using two IR mosfets, the whole thing running from a 500V rail (= 2000Vpp prim, 12000Vpp sec).
It was designed for a bandwidth of 100-20000hz. It works, but it doesn't help at al solving the common esl-transformer problems so I don't see the advantages over a normal off-the-shell amp driving a 1:100. So why go through the troubles of a 500V amp..

Additional problem I ran into is that you'll need to drive the transformer from a very low source impedance to maintain high freq response. This kind of output stage proved not very good at that, although I managed to solve it to satisfaction using lots of local feedback.

The big problem with transformers is octaves of bandwidth.
in combination with required output voltage, yes.

As an alternative I've been thinking about using two seperate transformers, splitting up the spectrum. One for 20-500, the other for 500-20000hz for instance.

Since I am not aiming for seperate bass/treble panels I want to add the outputs of the two transfo's and probably use some feedback to smoothen out the phase / amplitude errors created in the splitting/adding process. Basic idea as in picture. Problem is to maintain stability in the feedback loop.
 

Attachments

  • splittransf.gif
    splittransf.gif
    3.1 KB · Views: 436
Ex-Moderator
Joined 2003
I take your point about the problems of a 500V amplifier vs 50V. 50V is more common, so the problems have been solved and optimised parts are readily available.

I'm curious as to what sort of an ESL makes you even consider the paralleled transformer solution. I think some valve bench oscillators used two output transformers (although they may have switched between them). Seems like hard work.
 
To be more precise:

I mean asymmetric PWM where sqare wave = maximum AC Voltage, narrow spike = small AC voltage and a rectifier for amplitude demodulation after the transformer. I don`t know whether the IML works that way, but I read that it uses a small transformer.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.