| Horizontal deflection (or high voltage type) transistors to directly drive ESLs? - Click HERE for Original Thread |
| bigwill |
| Just wondering if you could do this without too much hassle. I don't know what else to ask really :) |
|
|
| Bazukaz |
I am currently working on this.
However , the max. voltage of these transistors is about 800-900V , and they are only available as NPN types.By using bridge , it is possible to get around 400-500 Vrms from this... Not so much , but should give some sound. |
|
|
| bobo1on1 |
| By using big inductors in a brigde amp the power supply voltage can be doubled, it will only work in class A though but class AB with only npn transistors is very hard anyway. |
|
|
| Bazukaz |
But , for as i understand , you cannot exceed transistors Vce rating , even if choke is used.In this case , only efficiency is incresed... By the way , could somebody explain why using a choke as a collector load does not cause full power bandwidth problems ?
I was thinking about quasi complementary AB output stage , but the PNP transistors , 2SA1968 , seems to be hardly avalable.
Regards,
Lukas |
|
|
| bobo1on1 |
You still need very high voltage transistors, fets or IGBT's might be even better since high voltage transistors usually have a low hfe.
Maybe current drive will also be better than voltage drive, pherhaps someone could elaborate on that. |
|
|
| Eva |
1200V IGBTs are rugged and quite cheap in comparison with MOSFET. Bridge output stages should be also considered, because 600V IGBTs are much more common (check IRG4BCxxx series from International rectifier and SKPxxNxx series from Infineon). For high voltage low-cost (switching) bipolar transistors check MJE1300x and MJE1800x series from On Semiconductor.
Series connecion of two or three lower voltage transistors with active dividers is another alternative, and of course, high efficiency class H is just an improvement over series connection. For example, consider an amplifier having +-1200V supplies tappered at 200V intervals (that may even come from a single pair of 200V rails with capacitor-diode multipliers), each interval having its own diode and output device (the problem here is to design a suitable VAS). |
|
|
| Bazukaz |
Hi,
Not necessarilry HV transistors have low hFE.Some have 30 or so.But their SOA is greatly decresasing with collector-emitter voltage.Because of this i am watching for fets now...
IRFBG20/30 1000V MOSFETS are cheap and easy to get.I am not sure if they are really suitable for audio , but , using a strong n/f, should work...
Regards,
Lukas. |
|
|
| Calvin |
Hi,
ther are a few High-Voltage-Trannies/-Modules with Uce in the kV-range.
Sensitron SCP-4979 is a module with 4 IGBTs, each 2.500V/1A
The Powerex CM200HG-130H is a single IGBT-Power Module with 6.500V/200A, the CM400HB-90H is a 4.500V/400A module.
Even though these are quite slow devices with high input-/output-capacitances, they should be fast enough for audio frequencies.
I don´t know anything about availibility and cost of these devices, but on the first glance they seem quite interesting fo me to replace the Pentodes in a ´classical´ High-voltage ESL-amp.
With such high Uces there should be no need for cascaded designs with their problems of generating the drive signals for the Bases/Gates.
Any comments?
jauuu
Calvin |
|
|
| rcavictim |
High voltage of many kilovolts and transistors. Which part of this suicidal comination don't you understand? :D
Think vacuum tubes here, enjoy reliability and never loose any sleep. |
|
|
| Bazukaz |
These IGBTs are amazing ... Ive just seen a datasheet of 6.5 kV 400A monster! Not so easy to destroy , even comparing with tubes :))).
I wonder about their pricing and availability ...Its probably worse than output transformer ... Isn't it ?
Regards,
Lukas. |
|
|
| Calvin |
Hi,
Transistors as well as Tubes have to be driven within their parameter´s limits. So why should a tranny be less reliable than a tube in a specific application??
A high- quality high-voltage amp for ESLs will be a inefficient design with quite strong heat development.
So it should be advantegous to use devices that don´t need heating at all.
A topology like a CCS loaded emitter-basis comes to my mind, eventually as a cascaded design (reducing the stress on the active devices), working in class A. Should allow for a very linear amp with sufficient current capabilities, without getting too inefficient.
jauu
Calvin |
|
|
| Brian Beck |
| quote: | | So it should be advantegous to use devices that don´t need heating at all. |
Unless one prefers the sound of tubes;) |
|
|
| Bazukaz |
Hi,
I have had no much experience with tube amps.But some blind tests reveal that testers were unable to identify good Tube amplifiers vs discrete ones(with one exception when amp is soft-overloaded).
So , my opinion , discrete approach is much more practical.Transistors don't need regular replacement , too.
Regards,
Lukas. |
|
|
| rcavictim |
| quote: | Originally posted by Bazukaz
Hi,
I have had no much experience with tube amps.But some blind tests reveal that testers were unable to identify good Tube amplifiers vs discrete ones(with one exception when amp is soft-overloaded).
So , my opinion , discrete approach is much more practical.Transistors don't need regular replacement , too.
Regards,
Lukas. |
Well then, the only best way to know for sure which is better for you is to build a set of each.
I am anoither tubophile like Brian. OOPs, sorry Brian, not implying you are tubby. I meant tube-o-phile. ;) Not only because I hear better sound from tubes but because I know tubes and high voltage are a better reliability mix Tubes won't go pop in a picosecond like most sand devices will if something ever goes slightly wrong.
I also live in a cold country so extra heat in my house is a bonus.
:D |
|
|
| Calvin |
Hi,
well, there are tubes and there are tubes. Having worked with KR-Audio I can honestly say..These babies are The Tubes. Nothing will top a KR-300Bxls or an T-1610...but guess what!! The driver stages in these exceptionally good sounding amps are FETs! Simply because they offer a superior signal handling in togetherness with an simpler and more straightforward design. Too, all these tubes are pure triodes. A lot of poweramp designs, especially the high voltage designs rely on pentodes. Signal handling of pentodes and MOS-FETs is very similar. So there is imo no reason at all not to opt for an simpler transistor design.
jauu
Calvin |
|
|
| Brian Beck |
| quote: | | Signal handling of pentodes and MOS-FETs is very similar. So there is imo no reason at all not to opt for an simpler transistor design. |
Except maybe that MOSFETs have a large and nasty non-linear input capacitance that pentodes (and triodes) don't have. Phase intermodulation distortion is not pretty. |
|
|
| rcavictim |
| quote: | Originally posted by Calvin
Hi,
well, there are tubes and there are tubes. Having worked with KR-Audio I can honestly say..These babies are The Tubes. Nothing will top a KR-300Bxls or an T-1610...but guess what!! The driver stages in these exceptionally good sounding amps are FETs!
jauu
Calvin |
Fets are the best the sand industry has been able to come up with to approach the superior audio performance of a good vacuum tube.
Perhaps KR skimped and used Fets as drivers because it they had put another custom tube in there nobody mortgaging just one house would be able to afford their amplifier.
KR is like a different planet when it comes to tubes. There was a very successful tube industry happening half a century ago without any influence from KR, and without the boutique prices. Fortunately one can have truly excellent tube audio without the KR expense.
This is only MY opionion. |
|
|
| Bazukaz |
Hi,
Again, i am posting the schematic of Neil S.Mckean with some questions:
What is the purpose to use Q1 to drive Q2 ? It cannot produce more current than LM318 does(around 20mA).
Another question is what C4,C5 and R11 does.It is kind of compensation ?
Thanks ,
Lukas. |
|
|
| Bazukaz |
| By the way , this is not a full schematic - the part that sets DC operating point is cut out. |
|
|
| Calvin |
Hi Victim,
don´t guess..read! "Simply because they offer a superior signal handling in togetherness with an simpler and more straightforward design." At KR they´ve tried out several tubed driver topolgies, but always came back to the MOSed driver stage because of its simplicity and superior performance!
jauu
Calvin |
|
|
| rcavictim |
| quote: | Originally posted by Calvin
Hi Victim,
don´t guess..read! "Simply because they offer a superior signal handling in togetherness with an simpler and more straightforward design." At KR they´ve tried out several tubed driver topolgies, but always came back to the MOSed driver stage because of its simplicity and superior performance!
jauu
Calvin |
Notwithsatanding my point that using another tube stage instead would have added a king's ransom or more to the amp price tag. |
|
|
| bobo1on1 |
What rms voltage would a direct drive esl amp need to produce to get a reasonable spl?
I was thinking about an amp working something like this: |
|
|
| Bazukaz |
Hi,
Lets say , we can archieve an output voltage of 500Vrms , by using bridge. in case a 50:1 trafo is used with a 100W amp , it would give around 1kV rms..
Thats ~6db loader than 500Vrms.
By using higher voltage transistors(1.5kV fets or IGBs are not too uncommon , though not as easy to get as 1kV ones), we may approach 800-900Vrms , which is close to a 50:1 trannie.
I am not sure about your schematic , i think it is not good at all.There is one at www.shackman.de and one that i posted ...
Regards,
Lukas. |
|
|
| bobo1on1 |
| I updated the schematic a little, R4 is now only used to set the dc current. |
|
|
| Bazukaz |
| DC operating point will vary with T1 hFe , won't it ? What about global feedback ? |
|
|
| bobo1on1 |
Those 3 transistors together should have a current gain of about 160.000, so unless you drive it with a very high impedance source there should be no problem at all.
Furthermore the amount of current needed to drive an esl is very low compared to a 8 ohm speaker.
Idle current will vary a little with T1's base-emitter voltage.
|
|
|
| Bazukaz |
| For drive current , we are talking about milliamps.Why not use an IC driver , and Power MOSFET output ? FETs are robust , while ICs are easy to work with...(unless you get them into oscillation :)) |
|
|
| Calvin |
Hi,
for the full power bandwidth to reach 20kHz and beyond, You´ll need a couple of hundreds of mA of drive current. Of course may You fudge a little with the needed bandwidth (there´s not much signal above 10kHz) to reduce the current demand, but for a full fledged solution with a high capacitance panel (metal sheets) You should calculate with 300 to 500mA of peak current.
A topology I´m thinking about is like the Beveridge or the Acoustat Amp but using transistors instead and a µ-follower like Allan Kimmel´s instead of the Schmitt/SRPP stages. This should provide for a very linear class A-stage without the need for feedback (feedback maybe just for positive DC-centering?).
jauu
Calvin |
|
|
| Bazukaz |
Well , for 300mA peak, and 1kV supply,
150W of idle power will be needed for one block.
If there are 2 bidged channels - thats 600W of steady power dissipation.A HUGE heatsink with fans , etc.
A dynamic tweeter , crossed at 2.5 kHz usually has a power rating of 1/10 of woofer's , or even less.
I dont think there is a need for so high current , unless panels are very highly capacitive.
Lukas. |
|
|
| bobo1on1 |
| Wat type of inductor should I use? I was thinking about the ones that are used to start TL's, they should be able to provide a decent voltage without arcing. |
|
|
| rcavictim |
| quote: | Originally posted by bobo1on1
Wat type of inductor should I use? I was thinking about the ones that are used to start TL's, they should be able to provide a decent voltage without arcing. |
Why do people constantly assume that everyone has the same brain wave patterns as themselves and understands their own pet acronyms sacrificing accurate communications of knowledge in the process?
On this forum the most likely meaning for the abbreviation "TL" would be "Transmission Line". They don't need starting with an inductor involving high voltage and steps to avoid arcing.
Could the poster please tell those of us here who are apparently ignorant, what the heck he means by "TL"? |
|
|
| bobo1on1 |
Sorry, by TL I mean tube light.
I ran a simulation using a resistor to keep the dc current accurate and a crossover to get a constant voltage above 10 hz or so, there is lots of oscillation, I don't think this is the way to go.
Also the panel and inductor will go in to resonance at a certain frequency.
btw the feedback is reversed in my previous schematics :clown: |
|
|
| Bazukaz |
Youp ,
I have already read it somewhere , a CCS is probably a better way to go...
Lukas. |
|
|
| bobo1on1 |
[s]I have done some simulations on a class A inductor amp, [s]it is hard to get stable and it has horrible clipping behaviour, I don't think this is the way to go.[/s]
Oops, made a mistake, used a 33k feedback resistor instead of 1M.
Anyway some steps should be taken to prevent oscillation. |
|
|
| Bazukaz |
A compensation cap of ~1-5 nF could be added across feedback resistor , to reduce gain at high frequencies.This might improve stability.
Also,the choke will generate twice the supply voltage , so FET will see it... I think so .
Lukas. |
|
|
| bobo1on1 |
After running some simulations I think the best thing would be two good tubes, two 60 henry inductors and a 600 volt power supply.
Some feedback is also needed to damp the resonance between the inductors and the esl.
I wonder how big that inductor would be, I think it will have to be huge. |
|
|
| Bazukaz |
Hi,
I was also thinking about the following.The cascode operation would allow for supply voltage of up to around 1.5kV by using 1kV FETs.A bridged version could give ~1kV RMS at output.
The problems would be limited power bandwidth and high power dissipation...
Regards,
Lukas. |
|
|
| bobo1on1 |
You need to connect R16 to gnd instead of the gate of U1, it now works like a weird kind of feedback.
Also 75k is a way too small value imho, you need at least 500k, of course this will limit the frequency response, so pherhaps some small caps in parallel to R16 and R22 would be wise. |
|
|
| Bazukaz |
What is the purpose to connect R16 to ground ?
I didn't invent this idea by myself , i found it at http://www.web-ee.com/primers/files/col_0412.pdf
The circuit shown is for JBTs , and i think it should be good for FETs too.
In simulations , it works well and shows f(-3) at > 100kHz.
Lukas. |
|
|
| Calvin |
Hi,
I find this amplifier very interesting. Its designed as an integrated IC-solution to drive capacitive loads.
Therefore CMOS and DMOS technology is used on one wafer.
It should be possible to adopt this structure for disrete designs with higher voltages and power.
The structure is as follows:
A low-voltage CMOS difference amplifier (3) driving an inverter (5). A current controlled current source CCCS (4) is connected in series with the inverter with regard to the high voltage supply- looks similar to SRPP and alike concepts. The following high-voltage part is fabricated in DMOS technology.
The inverter output is connected to the input of a sourcefollower (6).
The SF is connected to the high voltage supply and the output of the circuit. A diode D1 is connected between the ouput of the circuit and the connection of inverter and CCCS to discharge the load capacitance when output signal level sinks.
jauu
Calvin |
|
|
| Calvin |
Hi again,
second part of the amp.
The input signal is fed to the noninverting input of the OPamp (3), a 2 stage design consisting of the transistors M1..5 and M21, M22 (M3..15 form a biasing network). The negative input will be connected to a feedback network. The output voltage (M22) drives the Gate of the inverter stage, consisting of XMH1, 112, RT1, 2, CT1, 2 and zener diode D2.
The CCCS is made up of XMH5, 6, RT5...8, and CT5, 6.
Both stages drive the sourfefollower consisting of XMH3, 4, RT3, 4, CT3,4 and zener diode D3.
Zener diode D1 dicharges the load capacitance when output signal drops.
The specialty is the use of cascaded MOSFets (depletion) throughout the 3 high-voltage stages and the accompanying RC-biasing networks.
Function:
With 0V at the NI-input and output of the OPamp (M22) the inverter tansistors XMH1 and 2 are open, the CCCS transistors XMH5 and 6 are shut.
A positive going signal at INP opens up M22 and shuts down XMH1, 2. The CCCS now supplies the current to charge the input capavitances for the transistors. The voltage level at point 7 rises and opens up XMH3, 4 of the sourcefollower. So does the voltage at the ouput of the circuit, thereby charging up the connected load capacitance. A external feedback network connected between the output and the INN input controls the amplification level.
When the input signal is falling the whole circuit works similar as to the positive going signal, but with diode D1 helping to discharge the load capacitance. The inverter opens up and the sourcefollower shuts down.
The use of RC-Biasing networks shall ensure equal voltage distribution at the gates of the transistors with fast transients regardless of tolerances within the biasing resistors and input capacitances of the gates.
jauu
Calvin |
|
|
| Bazukaz |
Hi,
Finally , i have built it.For now , i have not tested THD and other parameters.Output is ~240Vrms at 350V supplies, but FETs are rated at 1kV so i think 800 supply should be no problem.
Regards,
Lukas. |
|
|
| bigwill |
| Hey! That's pretty cool! Does it work? |
|
|
| Bazukaz |
Yes , it works , but for now , i can only measure , but not listen , as ESL panels are waiting to be built , too :).
Regards,
Lukas. |
|
|
| Bazukaz |
Hi,
Finally , i set up my amp with esl to listen.The panel is 50x12.5 cm , 1.5 mm spacers , 3.8mm diapraghm , coated with thin layer of shoe polish.It fully charges in about 10-30 seconds.
EHT supply is approx 2.5 kV.
Amp's power supply is ~ 650V.That gives around 450Vrms at output.
In no way is it quet ! Panel's resonance is a bit too low , so it touches stators at full volume.Cutting out low frequencies with equaliser greatly improves maximum SPL.
However , the sound is a bit noisy, especially high frerquencies.
It seems that noise comes from panel , and not amplifier , but will have to check this out.
Anyway , its amazing to see such a thing producing sound !
regards,
Lukas. |
|
|
| bobo1on1 |
| What's your idle current? |
|
|
| Bazukaz |
Hi,
Approx 25-30 mA.
Regards,
Lukas. |
|
|
| bobo1on1 |
| Did you measure the amplifiers output? The 13k resistor and the panel work together like a lowpass filter, this could cause distortion at high frequencies. |
|
|
| Bazukaz |
Hi,
Output at 20 khz is less that 1 db down.
In this design , resistor limits full power bandwidth.At 200 pF load , it is approx 20kHz.
I was also thinking that internal mosfet's diode may open due to EHT supply leakage.EHT voltage is negative in respect to ground.
I will try to do some measurements when i get stators powder coated.
Regards,
Lukas. |
|
|
| bigwill |
| Would you say this design would be difficult to build for a newbie? |
|
|
| bobo1on1 |
| quote: | Originally posted by bigwill
Would you say this design would be difficult to build for a newbie? |
It's not that hard to build, just be very very carefull with the high voltage power supply.
ALso you need a polarizing supply just like when using a transformer. |
|
|
| Bazukaz |
Hi,
Really , it is not very hard to build.Still , the design is not finished yet.
While i havent got it working well , i cannot recommend to anybody.
Also , the high voltage middle point is grounded.So , protective grilles are required , so nobody can touch even one of stators.
Regards,
Lukas. |
|
|
| Bazukaz |
Hi,
Finally , i have made some measurements of the amp.
In simulations , everything worked perfect , with distortion levels <-80db between 20 - 10 000 Hz , at full output swings.
But , when measured , i get very high distortion(1%-15%) at low frequencies.Can this be "motorboating" ?
The most distortion comes in range of 20-200 Hz.
Also , rising output level greatly increases THD.
Regards,
Lukas. |
|
|
| bobo1on1 |
| Can you post an oscilloscope pic of the amp while its distorting? |
|
|
| Bazukaz |
Hi ,
Here is the distorting waveform.Does not look like clipping at all...
Regards,
Lukas. |
|
|
| bobo1on1 |
Is that the output of the two amps together working as a bridge or is it just one amp?
The dc offset might be wrong, causing one amp to clip before the other, this causes the gain to be cut in half, showing this kind of distortion. |
|
|
| Bazukaz |
Hi,
I measured with both amps connected , but only from one of outs to ground(I can measure only if input and output of my sound card is referenced to the same ground).
Here is circuit layout, maybe there is something wrong with grounding ?
Main feedback resistor's parallel capacitor is not included in PCB layout.
Also , connection between HV and LV grounds must be made with wire.
Driver ICs are NE5532.
Regards,
Lukas. |
|
|
| bobo1on1 |
| The weird thing is, the amplifier's topology is asymmetrical, while the distortion is symmetrical, its almost like the problem lies in the opamp or in the preamp. |
|
|
| maudio |
Hi,
I just joined in here and found this interesting tread. Maybe I can contribute here, I have quite some experience with direct drive amps for esl's. In the last couple of years I build several of them, ranging from 1kV single ended versions up to a 16 mosfet bridge version running on 4kV that puts out over 7000Vtt in a large fullrange panel. All using mosfets btw. They put out enough heat without tubes..
To sum it up briefly: Can direct drive work? yes. Does it sound better than a good stepup transformer? No. At least not if you want to drive large fullrange panels. However, for headphone use or for hybrid systems with esl mid/highrange, direct drive is very well possible.
The reasons behind are a long story, if anyone is interested pleae ask..
It all comes down to the need for AB operation due to dissipation problems and the impossibility to build a linear output stage using only N-devices (or tubes, for that matter). Remember, large esl's require >>5kV and peak currents up to 500 mA.
As for the distortion problem Lukas has, some suggestions that might help:
With low frequency distortion in the 20-200hz range I would at first suspect power supply (unsufficient buffering/poor load regulation). Try monitoring the power supply ripple with a scope, maybe add some more buffering capacity.
Another possibility would be ground loops or ground potential differences between your amp and your measuring device (=PC). Is the PC connected to grounded mains? If not, the RF filters in the PC's power supply can cause a lot of strange things to happen. Grounding problems in your pcb design will probably lead to hf instability instead of lf.
Some more suggestions for your design:
You use single resistors for R2-4-10-12, that won't work well unless you use special hv-types. Better to compose these resistors from several ones in series to keep voltage drop within safe range. Resistors subjected to more than about 250V will behave very non-linear, causing massive distortion (not to mention safety issues). Avoid carbon types, these are horrible onder high voltage conditions.
You drive the fets directly from the opamp output. In my experience that won't work very well, due to the miller (gate-to-drain) capacity in the fet that reverses large charges to the opamp output. Most opamps don't handle that very well, causing severe oscillation. Try increasing the gate resistor to at least 470R or even better, add a buffer stage (emitter follower). You might also want to replace the fets with IRFBG 20 (rated @1.4A it's powerfull enough and has much lower capacitances).
Your openloopgain is probably way too large and I am almost sure that this does cause instability. I suggest you check the gate signal for oscillations, that's the best place to see them. You'll need a real scope rather than a soundcard because oscillation will occur way beyond the frequency range of your soundcard (probably in the 500Khz-1Mhz range).
To stabilize the circuit, try lowering the loopgain by reducing the opamp's gain to around 10. That's enough for reasonable thd figures. In fact I never use opamps in my amps for this reason but I build most gain stages with single transistors.
By the way, replacing the drain resistors with current sources will double efficiency and is well worth it. You'll need the extra current it provides badly.
I must have a working design for a simply 2kv output HV amp using this topology somewhere, if you want to I can post it.
Hope my suggestions are of some help, good luck and be careful! |
|
|
| maudio |
One more remark:
spice simulations of such circuits tend to be very unreliable in my experience:
- spice modeling of mosfets is in general notoriously inaccurate,
- spice doesn't account for typical high-voltage related problems: high voltages create strong electrical fields that tend to couple weird signals into all places where you don't want/expect them, causing all kinds of unexplainable effects. There's also the fact that a lot of components deviate from their normal behaviour under hv-stress. To be accurate, you would have to use an elaborate model for each component, even the simple R's and C's.
Just something to keep in mind...
If it works in spice, it may or may not work in real life. On the other hand, if it doesn't work in spice you can be pretty sure that it won't work either in the real world.
regards,
martin |
|
|
| invisible force |
Hello Martin,
this thread is interesting, indeed.I can strongly support your comments on pspice and MOSFET´s behaviour. The crux of the problem is the voltage dependancy of any semiconductor capacity (not only for HV circuits).
Has any one ever thought of dielectric absorbtion effects
inside a MOSFET ?
<The reasons behind are a long story, if anyone is interested pleae ask. >
I am interested. Respect for your 16 MOSFET amp, simply for the existance..
Do you mind sending me the circuit for discussion?
Having wound countless transformers for ESL´s I would state : The amplifier/transformer combination is the limiting factor regarding sound quality, therefore direct drive is the future here.
If you take a decent STAX earspeaker and connect it to a very good transformer / low voltage amplifier and compare it to a SRM-T1 direct drive (by the way : a current source instead of the anode resistors has proven to be a good idea for the SRM-T1) it is clear, what is better.
regards,
Philipp |
|
|
| Bazukaz |
Hi,
maudio :
I am very interested in building a direct drive amp.I was able to reduce distortion of previous amp to <0,5% at 100Hz (0,07% @ 1k).I think that 2Watt carbon resistors(750Volt type) may be the problem, because distortion is higher at lower frequencies.
I plan to drive ESLs that are quite sensitive.I think that 800V psu with bridged outputs should work , at least for a first project.
Here :
http://www.diyaudio.com/forums/show...4846#post904846
I posted another schematic i have drawn; but even in Spice i can not stabilize it.I am not very experienced at this :((.
I would be great if you could share one of your schematics , not too complicated , if you have one :).
Best regards,
Lukas. |
|
|
| maudio |
Hi Phillipp and Lukas,
Phillip: I will answer your questions, but allow for some time... I also have to dig up the schematics, I build it a few years ago.
But I do include here the schematic of the 1kv design. It should work fine without too much trouble and the schematic is probably a good base for further discussions.
Some background information: I actually never designed this amp with the purpose of using it for driving my esl's, I rather built it to have a scaled-down version of bigger things to come without the additional problems of stacked output devices. It allowed for easy experimentation with different driver/feedback/etc configurations. But the end result is not bad at all for actually driving small esls or headphones.
The amp is a bridged single-ended class A design with current source loading. The 1 kV supply limitation is to avoid stacked fets.
Since there are no P-fets available for 1kv, I had to use N-types for the current sources. Which isn't the most elegant solution, the current needed for the voltage reference now flows into the output of the source. But by using leds as zeners I was able to reduce this current to 20uA so magnitude of this error is small. A current source built around a single mosfet this way isn't ideal anyway, but for this purpose it will do just fine. Amazingly this current source even turned out to be pretty stable, I never saw it oscillate.
As you see I don't use opamps but a current source loaded bc550 transistor as pre-stage. Together with the output stage gain there's plenty to achieve good thd figures.
Opamps add far too much gain here, to achieve good stability you have to reduce their gain in such a way that there is no point in using them anyway.
Performance:
Max output voltage swing is over 1900Vtt when ran from a 1kv supply. Output current is limited to 12 mA (quiescent current setting). THD is well below 0.1%, if I remember well I measured around 0.02% @ 1khz and around 0.08 @ 20khz between one output and gnd. So differentially (over the esl) distortion will probably be even lower. Don't expect to get much better numbers than this from any hv mosfet design (by the way Lukas, isn't the rise in low-freq distortion caused by your measuring software? Try increasing the fft-size).
Frequency range extends from a few hz (determined by C2) to around 60 khz (set by c9,c10 and the output devices internal capacitances).
The 12 mA output current will be sufficient only for a very small panel or headphone and the amp distorts heavily on larger panels due to current clipping. It is possible however to increase the current output (by lowering R3-12-4-10), if you also increase the power supply accordingly and provide enough heatsinking.
The practical limit of the design is around 40mA quiescent current, which will give a dissipation of 20W/fet totalling 80W/channel. This requires serious heatsinking (4 heatsinks each rated at < 1 C/W).
One way to get even more current would be to use the voltage drop over a resitor carrying the output current to modulate the current source (a bit like a SRPP) but it is very difficult to get this to work properly with a reactive load, let alone with complex loads such as a segmented wire-esl. Even if you do succeed in getting it to work, it will still only work for one very specific load. My experiences with it have been rather disappointing. You'll probably end up wasting a lot of extra heat in the output stage without getting much in return.
When posting a schematic like this I feel obliged to throw in some safety warnings:
- This is obvious NOT a newbies project nor a project for anyone without lots of experience in electronics, it uses lethal voltages and it CAN kill you. Use at your own risk and please take great care.
- Don't change anything or move measuring probes around while the circuit is live, use proper measurement gear rated for the voltages present. A 1:100 HV scope probe is well worth the investment.
- Each fet MUST have it's own heatsink. Mica of silicon insulation washers will fail sooner or later. Putting all fets on one heatsink might also give stability issues due to capacitive coupling between the fets, so don't.
- All heatsinks MUST be placed completely within the casing for safety reasons, no thermal washers will provide enough insulation to make the heatsinks safe to touch. You might want to consider forced cooling at higher currents. Cheap CPU coolers work excellent here, albeit a bit noisy.
- Be aware that this design puts a DC output voltage of 500V on the stators. For practical uses I would suggest adding 2 coupling capacitors with resistors to ground at the output, 10nf/2kv and 5*10meg in series should work fine.
This design shows in my opinion about the practical limits for what can be achieved using class A direct drive, in terms of output drive capabilities and dissipation. Disposing of any more heat in a safe way is impractical, especially since everything will have to be integrated with the esl panel, both for safety and to avoid EMC related trouble with your neighbours...
Looking forward to any reactions...
Martin |
|
|
| Bazukaz |
Hi,
Thanks for reply.
I am waiting for your schematic :)).
By the way , i think that it would be possible to use thick mica or silicone insulators , and ground the entire heat sink to improve safety.
I think that this method should give reliable and simple solution.
regards,
Lukas. |
|
|
| maudio |
| somehow the attachment was lost during preview, new try |
|
|
| Bazukaz |
Hi,
Thanks for schematic.Don't you think that 1kV supply is a bit too much for a 1kV mosfet ?As i remember , it has a temperature voltage derating , also , 10-15% mains fluctuations are possible ...
Maybe you got PCB drawings too ?
Best regards,
Lukas. |
|
|
| Netlist |
| quote: | Originally posted by maudio
- Each fet MUST have it's own heatsink. Mica of silicon insulation washers will fail sooner or later. Putting all fets on one heatsink might also give stability issues due to capacitive coupling between the fets, so don't.
|
Would it be possible to use ceramic insulators as shown
here ?
Perhaps the thermal resistance is too high.
I recall building an amp where they were mandatory with regard to the capacitance.
/Hugo |
|
|
| maudio |
| quote: |
By the way , i think that it would be possible to use thick mica or silicone insulators , and ground the entire heat sink to improve safety.
I think that this method should give reliable and simple solution.
|
Grounding the heatsink would prevent it from becoming live, but believe me, washers will most likely fail. Apart from sounding good you want your amp to be reliable as well...
The world isn't a perfect place and mica washers will have some microscopic pinholes, silicone washers when squeezed between heatsink and fet will have a spot a fraction of it's normal thickness (probably around the screw). All resulting in breakdown sooner or later.
The only option that will work is to use special ceramic washers, they do exist and are several mm thick (reducing capacity!) while still maintaining good heat conductance. Rated at 10kv or so. But they are hard to get and expensive.
Also remember that 4 small heatsinks can be smaller than 1 large because of better heat conduction to the air.
Using 1 kV on a irfbg20 is no problem at all, my experience is that these devices are very conservative rated and will withstand over 1200 V without any problems.
In fact, I never ever blew one that way. The only time I did destroy one was by letting it fry... (overheating) :D |
|
|
| Netlist |
I never had a question answered this fast. :)
/Hugo |
|
|
| Bazukaz |
Hi,
I liked the idea to use PVC wire to make small caps very much :).
One question : why did you use 4x LEDs in series instead of a zener diode ?
Also , did you make any considerations to stop electrostatic fields from influencing signal paths ?
Lukas. |
|
|
| invisible force |
Hi Martin,
Why not adding a source follower to the output - if it`s gate source connection is paralleled with a diode and kathode showing to the output Q1 and Q4 will sink the output current and the followers will provide the ESL with charging current. Idle current can be reduced significantly - less heat..Peak output current could be quite high by doing so.
regards,
Philipp |
|
|
| maudio |
| quote: | Originally posted by Bazukaz
Hi,
I liked the idea to use PVC wire to make small caps very much :).
One question : why did you use 4x LEDs in series instead of a zener diode ?
Also , did you make any considerations to stop electrostatic fields from influencing signal paths ?
Lukas. |
One has to be creative... You can also make good HV-c's from double sided printed circuit board.
I use leds because leds make a stable reference at much lower currents than zeners. The leds run at 20 uA, a zener wouldn't work at all at such low current.
I did not shield anything in this circuit because it wasn't necessary, I can make a picture since the amplifier is still lying around here. But you're right, it's not a bad idea to use shielding, in fact with my 4000V designs it proved to be vital.
One more remark about your pcb design: you should move the gate-resistors as close as possible to the fets. When they are a few cm away they won't be able to do their jobs (which is to prevent parasatic oscillations). And no, I don't have a pcb design for my amp, I build it on breadboard.. I'll post a picture. |
|
|
| maudio |
So this is what it lookes like...
Oh and I found an error in the schematic, please add 12V zeners between gate and source of the two bottom FETS for protection, will make them last a lot longer :)
I probably removed them from the schematic somewhere along the road because they tend to give troubles when running simulations, but you do need them. |
|
|
| maudio |
| quote: | Originally posted by invisible force
Hi Martin,
Why not adding a source follower to the output - if it`s gate source connection is paralleled with a diode and kathode showing to the output Q1 and Q4 will sink the output current and the followers will provide the ESL with charging current. Idle current can be reduced significantly - less heat..Peak output current could be quite high by doing so.
regards,
Philipp |
Added a picture, makes it easier to explain:
Yes, that is in fact the only way to do it and I used this in my 4000V amp. When stacking 8 fets you can even do some clever things combining resistors. But there's a catch:
This concept will work fine while running in class A, quiescent current being determined by R1 and R2 (10 mA in each fet here). But now imagine what happens when the circuit has to sink more current than R2 can handle: This current has to be sinked by Q1 through D1 and R3. This means that the voltage over D1 has to make an immediate change from about +3V (Ugs for the Q2 when conducting) to -0.7V, requiring the gate capacitance to change charge very rapidly. Feedback will try to achieve this by putting a very large spike on the gate signal of Q1, probably causing overshoot unless everything is very carefully compensated and nothing saturates. The scope photograph shows what happens if you don't get it right.
And there's more trouble: when Q1 draws current from the load, Q2 stops conducting altogether since it is pinched of by negative gate voltage. That causes a considerable change in the impedance 'seen' by Q1, which in turn causes a large change in total gain causing even more distortion. What's worse, the behaviour of the output stage in the frequency domain changes along opening the road for instability.
Add to the bad news list that the fets are not so linear anyway in this application, probably because of the low currents in the mA range. Even in open loop such a stage build from 8 fets has about 20% thd running in class A. When it enters class B everything goes off scale and what comes out has very little to do with what goes in. This concept is excellent for digital driver application but not so ideal for linear use. But, it's the only way to build something using only N-channel fets.
So what can we do about it with some feedback: In fact, you have to deal with 2 output stages having very different parameters since the output stage is constantly switching between 2 very different conditions. This makes compensation awkward en puts limitations on the amount of feedback that can be applied since the circuit has to be stable in both situations.
All in all, not exactly a good base for a high-end amplifier.. It can be done, I reached thd figures of around 0.1-0.2 %, but it simply doesn't sound good. Especially in the midrange (where esl's need most current) it sounds harsh and I don't like it.
This concept also has the issue of having 1/2 Vcc on the outputs which for a practical amp for use in you living room is simply unacceptable. This can be solved with coupling capacitors but when these have to be rated at over 4000V things start to get tricky...
Hope you start to get the picture why it's not so easy to build a good direct drive :) |
|
|
| invisible force |
Hi Martin,
< When stacking 8 fets you can even do some clever things combining resistors >
What you mean by this ?
This is really brainfood...open loop distortion has to be lowered some how...
Till soon,
Philipp |
|
|
| invisible force |
Hi Martin,
suggestion :
D1 should be held conductive by a voltage source (Vbe multiplier ...) inserted between Drain of Q1and R1, D1`s kathode direct to drain of Q1, R2 removed.
The idea behind :
As the load is almost a pure capacity there should be no current for a moment when derivation of output is changing sign ...
If D1 could be made conductive at that moment threre should be no nasty current peak because of sudden charge of Q2´s gate source capacity..
Am I wrong ?
regards,
Philipp |
|
|
| maudio |
For the brave-at-heart:
found a schematic of one of the bigger ones I built, do not try this at home :) It even produces soft music without the esl attached, I think because of ionization of the air between the heatsinks. Yes, that's scary.
I chosed to post this version because it shows all components seperate and includes the power supply. All others are more ore less simplified but this one gives some impression of the component count with such a design. How crazy can you be...
As I remember well, this specific one was optimized for maximum feedback in a futile attempt to minimize distortion. C8-9 are there to cancel out the first pole from the output stage allowing for some more feedback. With this I ran into the problem I described earlier, it is very tricky to keep things stable when the upper fets switch off and openloop parameters change. Stability margins were so narrow that it would start to oscillate when the load changed or even when it warmed up.
However, all versions use the same output configuration, differences were only in the pre-stages. You can built a pretty stable working version using a pre-stage similair to the one in the 1kV amp I posted earlier. But as I mentioned earlier, it doesn't sound good so don't bother. Apart from that, it's pretty lethal.
Phillip, I'll have to think about your suggestion, don't have much time now. Another solution I tried was to bypass D1 with a large capacitor in series with a small resistor. That works more or less (more less than more) but restricts bandwidth.
all good suggestions welcome!
have fun,
Martin |
|
|
| bigwill |
| quote: | Originally posted by maudio
For the brave-at-heart:
found a schematic of one of the bigger ones I built, do not try this at home :) It even produces soft music without the esl attached, I think because of ionization of the air between the heatsinks. Yes, that's scary. |
I'm hoping to build a loudspeaker based on this concept :D |
|
|
| Bazukaz |
Hi,
Once i have had an idea to use class D ic amp and high frequency toroid as a step - up.I simply connected little ferrite toroid to amp's outputs , without a filter , with series resistor for protection.But i didn't get it work this way(saturating transformer ?).
Any suggestions ? |
|
|
| invisible force |
hello,
I heard of this idea too much , transformer simpliy will saturarte instantly - because audio spectrum is inccluded to transform....
forget it,
Philipp |
|
|
| maudio |
I fully agree with Philipp on this.
And there's another problem with switching amps for esl: What do you think will happen in terms of radio interference when you connect an oscillator running at several 100 khz, producing steep flanks thus putting huge amounts of energy in high harmonics, with an output swing of several kV, to a very large antenna (your esl)...
I believe this is quite similair to the way Marconi bridged the Channel for the first time using radiowaves :D |
|
|
| Bazukaz |
Hi maudio,
I see you are experienced at this so i think you can help me a bit :).
I have rebuilt my schematic.Instead of a resistor load , i used CCS.Bias is ~45mA. Also , i increased closed loop gain of the amp .It became by far easier to stabilise.
I observed really strange things happening with THD curves.
With resistor divinder to sound card = 200k/0.4k , full output swing :
At 10K , THD is only 0.003% !!!
At 1K , THD is 0.4 % !!
if divider is changed to 1Meg/0.4K :
At 10 K THD = 0.03 %
At 1K THD = 0.08 %
At 100Hz Thd = 0.3 %
For 200K i used 2 5w wirewounds in series , for 1Meg - 2Watt carbon one.Wirewound didn't limit frequency response at all.
If it was a problem with opamps incapable of driving MOSFETs , the THD would rise at high freqs , wouldn't it ?
Also i observed differences (rising distortion) , if i short the source resistor of Q1.
What can be happening ? Problems with rf filters in PC power supply , as you mentioned ?
Regards,
Lukas. |
|
|
| maudio |
Hi Lukas,
<I observed really strange things happening with THD curves.>
The problem is in your soundcard, not in your amp.
Your soundcard is probably sampling at 44.1 or 48 khz. So it won't measure anything at all above 22 or 24 khz.
When measuring thd @ 1khz, your measurement will only include 2nd to 5th harmonics since the 6th is already at 32khz.
Similair, at 10 khz you're only measuring second harmonic distortion, all higher harmonics are outside frequency range of the soundcard.
So, the fact that you measure more thd @ 100hz than at 1khz may indicate that your amp produces a lot of higher harmonics (which is not good). But it may as well be that thd actually increases at low frequencies due to power supply noise (are you measuring thd or thd+noise).
The difference in results with the other divider are probably because the lower signal to the soundcard, increasing noise contribution.
Another consequence is that by using only soundcard, you can't be sure whether it's really stable or not since oscillations will be at frequencies way beyond what can be detected. But if you don't have an oscilloscope you can do crude oscillation detection by placing an AM-radioreceiver near the amp ;)
<For 200K i used 2 5w wirewounds in series , for 1Meg - 2Watt carbon one.Wirewound didn't limit frequency response at all.>
The problem with high voltage over resistors is not in frequency limits but in non-linearity. A 1:100 divider will be 1:100 at 100V but this can easily change 20 or 30% at 1kv. That is, when using carbon oor metal film. Wire wounds probably won't have this problem but they do have a lot of inductance.
<Also i observed differences (rising distortion) , if i short the source resistor of Q1.>
That's because the source resistor provides local feedback linearizing the mosfet (the fet is controlled by Ugate-Usource, over the source resistor there is a signal proportional to output current, thus part of the output signal is subtracted from the input signal). So, when you add a source resistor the gain of the fetstage decreases but it becomes much more linear.
This tells you that most distortion is produced in the fet (not surprising, fets are very non-linear): Adding local feedback decreases total openloop gain so it also lowers the total amount of feedback. But obviously, local feedback around the fet pays off more than is lost by decrease of global feedback.
In general, using local feedback is a good way to lower distortion without running into the stability problems when using hig global feedback.
Unfortunately, with the source-follower configuration I used in the 4kV design it's impossible to use local feedback by adding a source R: the current coming from the source is not proportional to output current anymore :bawling:
See that you can learn a lot about your creation even with a limited soundcard measurement system :D
Martin
|
|
|
| Bazukaz |
Hi,
The amp produces mainly 2nd harmonic.
Also there is a lot of noise , mainly 50 and 150 hz.I think it picks up mains hum somehow.
Also , i managed to blow one FET :cannotbe:
Regards,
Lukas
|
|
|
| maudio |
| quote: | Originally posted by invisible force
Hi Martin,
suggestion :
D1 should be held conductive by a voltage source (Vbe multiplier ...) inserted between Drain of Q1and R1, D1`s kathode direct to drain of Q1, R2 removed.
The idea behind :
As the load is almost a pure capacity there should be no current for a moment when derivation of output is changing sign ...
If D1 could be made conductive at that moment threre should be no nasty current peak because of sudden charge of Q2´s gate source capacity..
Am I wrong ?
regards,
Philipp |
Hi Philipp,
I hope I understood your suggestion well. I don't think it will work. The problem doesn't occur when the output current derivation changes sign, but when output current crosses the idle current setting. So, it will shift along the slope of the signal along with amplitude.
As I told before, I tried to reduce the problem by adding 220 u in series with a few hundred ohms between drain of Q1 and source of Q2. The idea behind is to create an alternatice pathway for the current, the C will maintain more or less Ugs level so the change in voltage over the diode will be much smaller. Since signal on the gate and source of Q2 is (almost) identical, it won't shortcurt the drive to Q2. Still, Q2 is turned of during part of each cycle and that is the main problem.
A better way would be to keep Q2 conducting all the time. I tried a thing or two but without much success before dropping this output configuration altogether, since will never meet safety demands anyway (drive voltage is relative to ground, so output will always carry 1/2 Vcc) so it will never result in a real practical amplifier.
Ideal would be a true push-pull output stage fed from symmetrical rail voltages: output at zero Vdc, low distortion. Among the things I considered to achieve this: stacking large amounts of low-voltage rated fets so P-channels come into view, or using optical drive to the output devices. But both have such practical problems that I never succeeded in getting it to work properly. Right now I am back at transformers and experimenting with enclosing them in the feedback loop.
But maybe someone comes up with a bright idea to overcome the difficulties with DD...
regards,
Martin |
|
|
| maudio |
| quote: | Originally posted by Bazukaz
Hi,
The amp produces mainly 2nd harmonic.
Also there is a lot of noise , mainly 50 and 150 hz.I think it picks up mains hum somehow.
Also , i managed to blow one FET :cannotbe:
Regards,
Lukas
|
Hi Lukas,
Indeed it looks mainly second... That's odd, by the way did you try measuring 'thd without noise'?
The only explanation I can come up with now for the strange thd curves is that your amp is oscillating. It would for sure explain the high figures you measure, with CCS load the fetstage will be quite linear and with an opamp in the path you have so much feedback applied that I'd expect thd to be in the thousands of a percent, if it was stable.
You write that it was easy to stabilize, but can you be sure... If I understand well, you don't have a scope available or am I wrong?
In fact, I am almost sure that your design can't be stable this way. Reason I am so sure is that I also started out using similair designs as you do now but I never got anything like it to work properly outside spice. They all proved impossble to stabilize.
You must lower openloopgain and you must add some frequency compensation (apart from c10). Skip the opamp, use an emittor follower to control the fet. It will work wonders.
Good luck and keep up the good work, much appriciation for your experiments!
Martin
(As long as there are more fets, there's hope) |
|
|
| Bazukaz |
Hi,
I have also experimented with this kind of circuit with BJTs as well.I have made a few experiments.Some of amps had the distortion , others not , with basically the same circuit.I haven't been able to detect the reason(once it dissapeared , by increasing supply caps ,but i wasn't able to repeat this).
I have came to several possibilities in my mind:
Problems with grounding ;
Oscillation;
Interference with sound card ;
Wrong main feedback resistor(2W carbon , 750V);
Interference with electrostatic fields created by relatively high voltages.
I have ordered some metal films , but i'll have to wait a couple of weeks until it arrives ...
I am not really sure about stability.It showed very good results in Spice , with very big tolerance to components.For sure , real world circuits may behave quite different , but i have found that spice is a good guide to find , what makes circuit more stable and what does not.
Thank you for helping me .
Best regards,
Lukas. |
|
|
| maudio |
| quote: | Originally posted by Bazukaz
Hi,
I have also experimented with this kind of circuit with BJTs as well.I have made a few experiments.Some of amps had the distortion , others not , with basically the same circuit.I haven't been able to detect the reason(once it dissapeared , by increasing supply caps ,but i wasn't able to repeat this). |
I think it is safe to assume that any such unpredictable behaviour is related to instability.
When using BJT's bear in mind safe operation area limits, a 1500V rated bu 508 for instance won't be any good over 700 V. That's why I use mosfets :D
| quote: |
I am not really sure about stability. It showed very good results in Spice , with very big tolerance to components.For sure , real world circuits may behave quite different , but i have found that spice is a good guide to find , what makes circuit more stable and what does not. |
Spice is quite reliable for normal small signal circuitry with bjt or opamps, but in my experience you can't rely on it at all with this kind of design. It seems to understand very little about mosfet behaviour under these conditions. Almost anything you try in spice will be stable but in the real world it's very different.
You must have some way to detect hf-instability, otherwise you basically have no idea at all what's going on. You can use a detection circuit placed before your DVM (see picture) to do some basic detection. Connect to your amp (through divider) with no input signal applied, any reading other than zero indicates presence of hf oscillations. But then again this still won't tell you whether your design remains stable under load and with signal, or if it shows ringing etc. To be honest, you can't do without an oscilloscope here.
| quote: |
Thank you for helping me .
Best regards,
Lukas. |
No problem!
By the way, do add some protection by means of a 12V zener between gate and source of all mosfets. Will prevent spikes from destroying them |
|
|
| Bazukaz |
Hi,
Thanks for schematic.I have read 5-20mV readings after the bridge , but , even with amp disconnected from mains , i still get 5 mV , so i assume it is the accuracy limit of my multimeter.
If i leave input floating or connect it to ground , DIM reads around 0.6V.
If i connect to sound card , i get ~20mV , or less sometimes.
One more thing - they use carbon composition resistors as distortion generators ! And they produce 2nd harmonic - the same as i get ! And , in my case,the distortion increases inversely proportionally to frequency !
Even if i use carbon films , it seems that these may give such an amounts of THD easily , keeping in mind high DC voltages across it :
http://www.geofex.com/Article_Folde.../carboncomp.htm
One more thing:
I tried to watch at music signal in spectrum analyser.It seems that
, at >10 kHz , there is almost 30-40db less audio content , than ~200Hz. So , an amp with full voltage output capability to around 5-7kHz may work without current clipping(i think).
Regards,
Lukas. |
|
|
| maudio |
| quote: | I have read 5-20mV readings after the bridge , but , even with amp disconnected from mains , i still get 5 mV , so i assume it is the accuracy limit of my multimeter.
If i leave input floating or connect it to ground , DIM reads around 0.6V. |
0.6 for sure means trouble. The other readings are strange, you are measuring DC I hope? Use a divider of 1:10, will give you much better signal to noise ratio when measuring.
Thanks for the link about resistor distortion, interesting article... I guess this is good news for the addicts of tube sound, a much cheaper way to produce their beloved harmonics :D
| quote: | I tried to watch at music signal in spectrum analyser.It seems that
, at >10 kHz , there is almost 30-40db less audio content , than ~200Hz. So , an amp with full voltage output capability to around 5-7kHz may work without current clipping(i think). |
You're right, but still you need around 50 mA to drive an ESL with 2000Vtt.
Regards,
Martin |
|
|
| Bazukaz |
Hi,
I think i have figured out the reason .. Sound card !
It is strange that is works well in loop configuration(out connected to in).
I set up Laptop PC for measurements , and all problems were gone.
THD+N fugure is now very realistic , showing 0.02% or less up to 7 kHz.At 1 kHz , THD+N is ~0.008%
IMD test results are similar.
Regards,
Lukas. |
|
|
| maudio |
Hi Lukas,
that's good news. Under what load did you do these measurements?
Next step to extend it to a bridged amp, hook up your favorite esl and test it, to check that it not only measures good, but also sounds good :)
(yes, those are two different things)
Good luck, keep us informed.. |
|
|
| Bazukaz |
Hi,
I measured it under 470pf load.I should have done with more , at least 1500pf , to get an idea how it works.By the way , IMD is around 0,015%. For comparison , LM3886 has only 0.004%. Is it bad , or shouldn't i care ? I guess it is the product of non-linear capacitance of FETs...
For now , i have problems obtaining mylar :(. I have made a wire ESL , and tried to use baking film , but is seems to be noisy ; PVC is by far too weak to hold any tension ...
www.audiocircuit.com sells mylar for 24$ , but only 10 metters - not so much for experiments ...
I have also drawn a PCB for this amp , it would be great to know what do you think about it.By the way , i included LEDs instead of zeners in CCS , and voltage regulated supply for op amps...
I hope to use mica insulators and avoid using standard screws with holes.The insulators are rated at 2kV.
Kind regards ,
Lukas. |
|
|
| maudio |
There seems to be a lot more on the pcb than in your last schematic... So could you post the schematic as well, makes it easier to understand layout...
I wouldn't care about distortion figures too mcuh. What matters is how it sounds. |
|
|
| Bazukaz |
Hi,
Here is the schematic.As i have told , zeners are replaced by 4 LEDs.
Schematic does not include Op Amp supply as well.
Regards,
Lukas. |
|
|
| maudio |
Hi Lukas,
looked at your latest design, looks ok, some suggestions to think about :
- Very important: good hf decoupling. C2,6,7 should be bypassed with some small foil capacitor (0.1 u mkt). Also, place a small (10nf) foil capacitor (2 kv) on the pcb decoupling the hv supply. The electrolytic buffer caps in the power supply won't do much at higher frequencies, increasing risk of oscillations.
- R8/13 must be rated 1000V, did you take that into account? They also can be of much higher value than 1 meg, I used 18 meg without problems. Will decrease dissipation and will also make the current source behave much better.
- With large r8/13, C6/7 can be much smaller (100uf or even less).
- same for c2, a few uf will do.
- Again,you must add protection diodes (12 v zener) between gate and source of u1/4, anode to source, will save them from gate isolation breakdown when you start feeding signals containing steep slopes (aka music).
- Why need c1/4? I think you can safely omit them. They only drain current and you'll need every mA you can squeeze out of it.
- I would increase R14/21 to 1 or 2 kohm, won't affect frequency range at all and will increase stability a lot.
By the way, 10 meters of mylar seems like a lot to me. What do you intend to build :bigeyes:
Your amp won't deliver enough voltage/current for a large panel anyway. I would suggest you build a small panel, maybe 15 x 100 cm surface area and also keep spacing from membrame to stator at <1.5 mm. Thighten membrame as much as possible, will allow for more bias thus better sensitivity. Otherwise your amp won't be able to drive it any louder than whispering levels... I know from experience. |
|
|
| Bazukaz |
Hi,
Thank you for suggestions.I will update the PCB and include HF decoupling caps.As far as i know , disc ceramics are the best for this purpose ...
One question : how is it possible , that FET gate could receive higher voltages , if op amp supply is only +-15 V ? IRFBG30 datasheet says , that max gate voltage is +-20V.By the way , i use these instead of IRFBG20 because of better availability in my city.
I included C1/C4 because it seems to stabilise circuit , at least in spice simulations.I could omit them , and rely on capacitance that will be formed between transistor and heatsink(grounded).If i would use mica , i would get ~140pF ; some thicker insulators would give less capacitance.
I have built a panel with capacitance of ~700 pF.(20x150cm).
I tried to simulate this load ; at 40 mA current , the amp can deliver
about 20%-25% less voltage(10kHz) , compared to 1 kHz. I think it should be enough , shoudn't it ?
I have also inspected the PCB and found a few errors.
Best regards,
Lukas. |
|
|
| maudio |
| quote: | Originally posted by Bazukaz
[B]Hi,
Thank you for suggestions.I will update the PCB and include HF decoupling caps.As far as i know , disc ceramics are the best for this purpose ...
One question : how is it possible , that FET gate could receive higher voltages , if op amp supply is only +-15 V ? IRFBG30 datasheet says , that max gate voltage is +-20V.By the way , i use these instead of IRFBG20 because of better availability in my city. |
The danger comes from the other side: there are internal capacitances present in the mosfet. The one between drain and gate (~50pf) will transfer charge from drain to gate (this is also why you'd be better off with a emittor follower between opamp and fet, will prevent the charge from interfering with internal feedback mechanisms in the opamp).
Steep transients on the drain cause large spikes on the gate that can easily destroy it (dV/dT over Crss). Such transients occur in music but will also be generated by spikes on mains or in case of the inevitable small static discharges within the esl. You have only been testing with sine waves and with a very clean load so far, real life is very different. If you have access to a scope, try what happens on a 1 khz square wave, you'll be surprised...
| quote: | | I included C1/C4 because it seems to stabilise circuit , at least in spice simulations.I could omit them , and rely on capacitance that will be formed between transistor and heatsink(grounded).If i would use mica , i would get ~140pF ; some thicker insulators would give less capacitance. |
.
It will probably stabilize the circuit by reducing bandwidth. That can also be achieved with a small c over the opamp, this solution won't cost you any output current.
Again, better use seperate heatsinks. Will save you 140 pf additional load that drains output current you badly need for your panel, and will also increase reliability a lot. I have bad experiences with combining mosfets on 1 heatsink. If you insist on 1 heatsink try to get the special ceramic ones as pointed out by Netlist earlier in this tread.
| quote: |
I have built a panel with capacitance of ~700 pF.(20x150cm).
I tried to simulate this load ; at 40 mA current , the amp can deliver
about 20%-25% less voltage(10kHz) , compared to 1 kHz. I think it should be enough , shoudn't it ? |
700pf is a lot to drive. But I suppose you'll be using segmentation? That will be necessary anyway, or it will bundle high frequencies too much.
Theoretically 40 mA may seem enough, but music isn't composed of sine waves. I measured peak stator currents up to 3 times what you'd expect based on capacitance, voltage and freq. Obviously there are more complex things going on here.
I think your panel is a bit large for this design, but again a lot depends on segmentation and your d/s spacing. You have to find an optimum, more d/s will provide more output SPL but will also require a lot more drive voltage (reduced sensitivity). What use is a 3 mm stroke if you can only drive it 0.2 mm...
So you have to match d/s to your drive voltage. In my experience 1 mm is about right for a 2kVtt amp. |
|
|
| Bazukaz |
Hi,
So , you really think that emitter follower would really improve things ?
Also , if i consider using it , i should add a diode from emitter to base , to protect it from op amp trying to drive it with negative voltages ?
One more : should i add zener from source to gate for CCS as well ?
About drive current :
Lets say , we have 4 ohm 100W amp with 10A current limit, driving ESL through 1:100 step-up.
So , max current at secondary is limited at 100mA. 1 nF load will look as 10 uF for amp; it has 1.6 ohm impedance at 10 kHz.So , I = 20/1.6 = 12.5A. We are already exceeding current limit of amp ! But it will be able to deliver much less than this , even much less that 10A , probably 2-3A in sine wave , because amp will have to dissipate all the power it delivers.
I made an assumption that amp characteristics are similar to LM3886.
Please correct , if i am wrong somewhere.
Best regards,
Lukas. |
|
|
| bobo1on1 |
| You don't need that much power at 10 kHz, unless you want to cause headaches. |
|
|
|
|
|
