Actually, I'm not interested at all in standard output transformers (though it would be interested to apply this around a Berning ZOTL stage, where he uses a high frequency ferrite transformer and a switcher for impedance conversion).
I'm more interested in applying it to two amplifiers that have gain in their output stages. They are both single-ended. One is the Aleph-X, an amp discussed much in this forum. The other is a hybrid amp for electrostatic headphones (link).
On a side note, I'm wondering about using triodes in place of Q13/14, though that would need positive feedback to get as much gain as the MOSFETs are providing.
One reason I'm interested in these two amps is that I'm building both; already got all the parts.
I'm more interested in applying it to two amplifiers that have gain in their output stages. They are both single-ended. One is the Aleph-X, an amp discussed much in this forum. The other is a hybrid amp for electrostatic headphones (link).
On a side note, I'm wondering about using triodes in place of Q13/14, though that would need positive feedback to get as much gain as the MOSFETs are providing.
One reason I'm interested in these two amps is that I'm building both; already got all the parts.
The electrostatic headphone project looks like a classical very high gain global NFB SS design with tubes subbed in for the final HV outputs. Q13 and Q14 are performing the voltage to current conversion for the grounded grid outputs. The "trioded" EL34 make no sense to me, they just force Q13 and Q14 to work harder. The tube puts out the same current it is driven by. Maybe screen current distortion was the concern. That could be fixed by using HV mosfets to drive the screens at constant voltage and their drains connected to the plates to recover the screen currents.
My first impression was that this design is hopelessly too high gain in the output stage to consider for an error correction scheme. Error correction is best left to heavily degenerated output stages where the gain varies very little. But there is some hope. First, the tube gm does not determine voltage gain here (Q13/14 do) so its drift with time will not destabilize the EC.
I would put current sensing resistors in the drains of Q8,Q15 for the EC feedback sampling. This will allow one to linearize the output current versus drive voltage at points A/B. Another diff ampl. stage would normally be required to form the EC signal. But with all the diff amps here, one could try to do this in Q13/Q14 by returning their sources to the sense resistors. Then some EC feedback resistors from Q13,Q14 drains to point A,B or equivalent. Need to get phasing correct in all this so error is corrected. Pots required to adjust the EC level. Too much EC turns it into an oscillator (well, the global feedback will normally prevent this except in a clipping state).
Couldn't find a schematic on the Aleph-X.
Don
My first impression was that this design is hopelessly too high gain in the output stage to consider for an error correction scheme. Error correction is best left to heavily degenerated output stages where the gain varies very little. But there is some hope. First, the tube gm does not determine voltage gain here (Q13/14 do) so its drift with time will not destabilize the EC.
I would put current sensing resistors in the drains of Q8,Q15 for the EC feedback sampling. This will allow one to linearize the output current versus drive voltage at points A/B. Another diff ampl. stage would normally be required to form the EC signal. But with all the diff amps here, one could try to do this in Q13/Q14 by returning their sources to the sense resistors. Then some EC feedback resistors from Q13,Q14 drains to point A,B or equivalent. Need to get phasing correct in all this so error is corrected. Pots required to adjust the EC level. Too much EC turns it into an oscillator (well, the global feedback will normally prevent this except in a clipping state).
Couldn't find a schematic on the Aleph-X.
Don
Open loop gain is 2000, with feedback 1000, so the feedback isn't that much. Given you need 800 V peak to peak output, that would mean it's driven to full power by a 0.8 V p-p input. So the gain is appropriate for the application.smoking-amp said:
Well, that's why there are the Q8 and Q15 followers between these and the tubes, no?
How does that work?
The text says Q13 and 14 are doing 200 gain. That's most of the gain in the circuit then. Wouldn't it follow that the output tubes are doing far less gain?
I'll put this in a simulator in the next couple of days and post again.
LOL, there's tons of discussion in diyaudio about it.
The most common schematic is http://www.diyaudio.com/forums/attachment.php?s=&postid=125874
I'm building a version with higher voltage and paralleled output devices for more power. I think THD is reasonably high in this design, maybe even 1%, so EC would be a benefit, whereas the electrostatic amp is supposedly 0.004% (who knows if that's with the capacitive load of the Stax plugged in).
Normally, grounded grid configurations are used to get very high gains, but it depends on the load impedance since they are just current output circuits. If the voltage gain really is only 2 here, then the triode Mu of the EL34 is not an issue, it will look like a pentode (with triode connection) for all practical purposes.
The MosFet drive for the screens would connect MosFet source to screen (thru a 100 Ohm damper resistor), the gate to a fixed screen voltage reference, and the drain to the plate. Any current flowing out of the screen gets passed thru to the drain and re-inserted into the plate current stream. However, with 800 V P-P output this could be a problem to find a suitable MosFet. And if you do find one, why not just delete the tube altogether?
My suggestion of using a modified Q13/Q14 to derive the EC signal isn't working unfortunately, it still is primarily outputting signal prop. to the input signal. Will require a separate diff. ampl. with resistor loads. Although using the sense resistor feedbacks in Q13,Q14 sources may still be a good idea for linearization as long as Q13,Q14 are not drawing any significant AC signal current thru the sense resistors.
Have to run, maybe can look at the Aleph-X later tonight.
Don
The MosFet drive for the screens would connect MosFet source to screen (thru a 100 Ohm damper resistor), the gate to a fixed screen voltage reference, and the drain to the plate. Any current flowing out of the screen gets passed thru to the drain and re-inserted into the plate current stream. However, with 800 V P-P output this could be a problem to find a suitable MosFet. And if you do find one, why not just delete the tube altogether?
My suggestion of using a modified Q13/Q14 to derive the EC signal isn't working unfortunately, it still is primarily outputting signal prop. to the input signal. Will require a separate diff. ampl. with resistor loads. Although using the sense resistor feedbacks in Q13,Q14 sources may still be a good idea for linearization as long as Q13,Q14 are not drawing any significant AC signal current thru the sense resistors.
Have to run, maybe can look at the Aleph-X later tonight.
Don
Actually the hybrid amp is an update to a completely solid state amp:
http://headwize.com/projects/showfile.php?file=gilmore2_prj.htm
From what I remember reading on a discussion forum, he found that the tubes' curves nulled those of the SS reasonably well; I forget which of the SS devices were the ones in question, the CCS transistors or the driving FETs), and that was one motivation.
As for why not use FET output... well this is a tube forum. I'd say, why not use more tubes? I thought CCS-loaded tubes are great voltage gain stages; why not replace Q13 and Q14 with triodes then. This stage swings under 100 V, and I've got some 6DJ8's I don't know what to do with... Though the tubes may need some Wolcott-style positive feedback to raise the gain to what these MOSFETs are doing.
BTW, the efficiency is pretty bad, and the 2SK1968 in the CCS actually overheated for some that built the amp, so now they use even more of them... I was wondering about improving efficiency by modulating the current source as done in the Aleph amps.
A few miscellaneous questions, btw, for anyone that can illuminate: How much of an improvement would driving the screens have? I've got some 1500 V transistors, though no MOSFETs near that.
Also, any benefit to increasing the degeneration resistors on the 2SK389 input? BTW, I'm not sure how the feedback resistors are calculated, since they don't go into the gates where the signal goes, but to the sources (the reasoning given was that feedback to gates would require very large resistances which have more parasitics, or something). And what do the 1N419 do? I thought diodes in the signal path was bad. Finally, someone mentioned that MAT04 transistors are better than the 2SK689 JFETs, is this true?
http://headwize.com/projects/showfile.php?file=gilmore2_prj.htm
From what I remember reading on a discussion forum, he found that the tubes' curves nulled those of the SS reasonably well; I forget which of the SS devices were the ones in question, the CCS transistors or the driving FETs), and that was one motivation.
As for why not use FET output... well this is a tube forum. I'd say, why not use more tubes? I thought CCS-loaded tubes are great voltage gain stages; why not replace Q13 and Q14 with triodes then. This stage swings under 100 V, and I've got some 6DJ8's I don't know what to do with... Though the tubes may need some Wolcott-style positive feedback to raise the gain to what these MOSFETs are doing.
BTW, the efficiency is pretty bad, and the 2SK1968 in the CCS actually overheated for some that built the amp, so now they use even more of them... I was wondering about improving efficiency by modulating the current source as done in the Aleph amps.
A few miscellaneous questions, btw, for anyone that can illuminate: How much of an improvement would driving the screens have? I've got some 1500 V transistors, though no MOSFETs near that.
Also, any benefit to increasing the degeneration resistors on the 2SK389 input? BTW, I'm not sure how the feedback resistors are calculated, since they don't go into the gates where the signal goes, but to the sources (the reasoning given was that feedback to gates would require very large resistances which have more parasitics, or something). And what do the 1N419 do? I thought diodes in the signal path was bad. Finally, someone mentioned that MAT04 transistors are better than the 2SK689 JFETs, is this true?
Hi Nixie,
Finally got around to looking at the Aleph-X schematic, my PKzip program expired, had to download Winzip. (why do people zip already compressed programs? .zip should be banned)
The Aleph-X is obviously a low feedback design, only two transistors in signal path Q5/7 and Q2/11, everything else is current sources or safe operating area protection. R46,R47 appear to be some tweek to optimize distortion at op. point, maybe by dist. cancellation between Q5/7 and Q2/11 pairs. The outputs are a significant part of the voltage gain with little degeneration. The characteristic curvature of the outputs will severely limit the amount of error correction that can be applied here before oscillation breaks out (and with minimal gNFB to control/prevent oscillation). Not really a good design to apply EC to.
(EC either needs strong global NFB to control osc. when passing thru >1 error corr. point, and/or needs to use some error feedforward so err. feedback can be dropped sufficiently below unity error corr. level )
An EC circuit would use another diffl. pair to compare output signals with Q2/Q11 drive signals (4 resistors, inverting null config.) Outputs would cap couple to the Q2/Q11 drives again. Use a matched pair of bipolars like Analog Devices Mat0x series, resistor degeneration in the emitters to get unity error correction gain (pots in parallel for tweeking gain).
Problem is, even if you get this to work for linearizing the outputs, the design was likely using complementary dist. from the input pair to correct that to begin with, so you will be left with the input pair contra dist. So you will then want to change the inputs to a MAT0x matched pair with some emitter degen, delete R46,R47.
You will be ending up with a totally different amplifier, and grounded source Q2/Q11 outputs are not good approach for EC. Better to use source followers, then you need more gain in the input stages, like maybe another diff. stage.....
Maybe could try applying EC to the whole amplifier instead of just the output Qs. Then could leave amplifier internals intact. But gain is even higher then, not good for EC, but depends on how stable that existing gain is, ie, how effective existing internal dist. corr. is. Also places more severe constraints on EC phase and amplifier phase at HF.
Any dist. data available for this design? Particularly versus signal amplitude.
Don
Finally got around to looking at the Aleph-X schematic, my PKzip program expired, had to download Winzip. (why do people zip already compressed programs? .zip should be banned)
The Aleph-X is obviously a low feedback design, only two transistors in signal path Q5/7 and Q2/11, everything else is current sources or safe operating area protection. R46,R47 appear to be some tweek to optimize distortion at op. point, maybe by dist. cancellation between Q5/7 and Q2/11 pairs. The outputs are a significant part of the voltage gain with little degeneration. The characteristic curvature of the outputs will severely limit the amount of error correction that can be applied here before oscillation breaks out (and with minimal gNFB to control/prevent oscillation). Not really a good design to apply EC to.
(EC either needs strong global NFB to control osc. when passing thru >1 error corr. point, and/or needs to use some error feedforward so err. feedback can be dropped sufficiently below unity error corr. level )
An EC circuit would use another diffl. pair to compare output signals with Q2/Q11 drive signals (4 resistors, inverting null config.) Outputs would cap couple to the Q2/Q11 drives again. Use a matched pair of bipolars like Analog Devices Mat0x series, resistor degeneration in the emitters to get unity error correction gain (pots in parallel for tweeking gain).
Problem is, even if you get this to work for linearizing the outputs, the design was likely using complementary dist. from the input pair to correct that to begin with, so you will be left with the input pair contra dist. So you will then want to change the inputs to a MAT0x matched pair with some emitter degen, delete R46,R47.
You will be ending up with a totally different amplifier, and grounded source Q2/Q11 outputs are not good approach for EC. Better to use source followers, then you need more gain in the input stages, like maybe another diff. stage.....
Maybe could try applying EC to the whole amplifier instead of just the output Qs. Then could leave amplifier internals intact. But gain is even higher then, not good for EC, but depends on how stable that existing gain is, ie, how effective existing internal dist. corr. is. Also places more severe constraints on EC phase and amplifier phase at HF.
Any dist. data available for this design? Particularly versus signal amplitude.
Don
Thanks for the reply. I'll try playing around in simulation.
You mentioned error feedforward; is this applicable to either of the two amps?
BTW, I want to ask about EC on a transconductance stage, where ouptut voltage is (near) constant. An example is plasma speakers like the Plasmatronics, where the discharge is constant voltage as current varies over the operating range (typical circuit is pentode or tetrode acting as voltage-controlled current sink hanging off the plasma cathode, with the supply voltage applied at the plasma anode).
You mentioned error feedforward; is this applicable to either of the two amps?
BTW, I want to ask about EC on a transconductance stage, where ouptut voltage is (near) constant. An example is plasma speakers like the Plasmatronics, where the discharge is constant voltage as current varies over the operating range (typical circuit is pentode or tetrode acting as voltage-controlled current sink hanging off the plasma cathode, with the supply voltage applied at the plasma anode).
EC for a transconductance stage would just require some current sense resistors for the output sense signal derivation to subtract from the drive signal to derive the EC feedback signal. Hopefully not requiring floating resistors. This will linearize output current with drive signal.
Feedforward is conceptually easy to add (see figure 1.1 in generalization paper, b path) but is a PITA to do practically usually. One needs another very linear output stage with comparable gain to the main output and some means to sum the outputs (usually series added).
There is a paper on EC applied to a current mirror, that uses something like feedforward in additionto the EC.
(Actually, it looks more like what I try to do in the next section->)
((Applying EC to a vac. tube current mirror with current gain is one of my projects on the back burner for the time being, mentioned in my thread on the vac. tube current mirror or "long lost linear gain stage".))
I plan to try a different scheme with pentode outputs where I will feed some additional EC signal to the screen grids of the actual output tubes. This would not technically be "feedforward" since it is injected before the output sampling point, it looks more like just some extra conventional NFB (and not EC either, since it is not affecting the input reference level, although this will be combined with an actual EC loop to the input as well). But I am hoping that this will still accomplish the same goal, to reduce the EC feedback correction requirement to below full unity error correction, so EC loop stability will be gained. Not sure if this is working yet, have to build or simulate. The screen grid input is at least fairly linear for a small correction level due to its low gm. Will be called EC + active ultralinear if all works out.
Don
Feedforward is conceptually easy to add (see figure 1.1 in generalization paper, b path) but is a PITA to do practically usually. One needs another very linear output stage with comparable gain to the main output and some means to sum the outputs (usually series added).
There is a paper on EC applied to a current mirror, that uses something like feedforward in additionto the EC.
(Actually, it looks more like what I try to do in the next section->)
((Applying EC to a vac. tube current mirror with current gain is one of my projects on the back burner for the time being, mentioned in my thread on the vac. tube current mirror or "long lost linear gain stage".))
I plan to try a different scheme with pentode outputs where I will feed some additional EC signal to the screen grids of the actual output tubes. This would not technically be "feedforward" since it is injected before the output sampling point, it looks more like just some extra conventional NFB (and not EC either, since it is not affecting the input reference level, although this will be combined with an actual EC loop to the input as well). But I am hoping that this will still accomplish the same goal, to reduce the EC feedback correction requirement to below full unity error correction, so EC loop stability will be gained. Not sure if this is working yet, have to build or simulate. The screen grid input is at least fairly linear for a small correction level due to its low gm. Will be called EC + active ultralinear if all works out.
Don
I'd be interested in the possibility you mentioned of applying EC to the whole Aleph-X instead of just the output stage.
Your mention of feedforward, what do you mean it's usually summed in series? I don't quite follow. The only passive summation I've seen is through resistor network.
Your mention of screen grids...
I intend to use 4X150A ceramic tetrodes for the plasma (it's what I've got, rare Eimac version with gold-plated control grids), and I've been wondering what to do with the screen grids for best results...
An unusual example of an output stage is the impedance conversion switcher in Berning's ZOTL (http://www.davidberning.com/patents.htm). There is actually a fractional voltage gain here. Is this too weird to have EC around it?
As for current mirrors, I remember Broskie wrote in an old TCJ where he claimed a current-mirrored stage is basically push-pull in nature, and I tend to prefer single ended (http://www.tubecad.com/march2001/page22.html right par.2).
Your mention of feedforward, what do you mean it's usually summed in series? I don't quite follow. The only passive summation I've seen is through resistor network.
Your mention of screen grids...
I intend to use 4X150A ceramic tetrodes for the plasma (it's what I've got, rare Eimac version with gold-plated control grids), and I've been wondering what to do with the screen grids for best results...
An unusual example of an output stage is the impedance conversion switcher in Berning's ZOTL (http://www.davidberning.com/patents.htm). There is actually a fractional voltage gain here. Is this too weird to have EC around it?
As for current mirrors, I remember Broskie wrote in an old TCJ where he claimed a current-mirrored stage is basically push-pull in nature, and I tend to prefer single ended (http://www.tubecad.com/march2001/page22.html right par.2).
For series summation, one can use a secondary winding in series with another output, or one can put the load "differentially" or "bridged" across the two outputs (other side of outputs go to ground, and phasing is such that the signals add in the load).
I would try any experiments with screen drive out on something cheaper first, before using the 4X150s. Gain some experience first. ( perhaps I should warn you that I ONLY experiment with new ideas, untread territory for tubes, I'm BORED with the usual stuff, so don't expect canned, proven solutions here.)
The Berning-SwitchMode OT should work for EC designs, however a few caveats apply. It requires a low pass filter to remove switching artifacts, so bandwidth is limited (as conventionally constructed anyway, I have a 4 phase version that does not have these bandwidth limitations due to the switching ) , but usually one would only consider applying EC before the OT anyway.
Next caveat is if you want to do partial CFB with the B-SMOT, the % cathode winding section also gets % fractional "B+" power supply voltage injected, so the cathode is not truly referenced to ground then. This can be made to work, but requires some extra measures. Partial CFB is very useful for pentode outputs and for a unity gain tap point for the EC.
You also have to deal with EMI filter/shielding issues for a conventionally configured B-SMOT. (the 4 phase version however does not generate any significant EMI since it uses controlled switching slew rates.)
Current mirrors conventionally added onto a diffl. amp. stage do indeed become a P-P design. However, current mirrors by themselves have nothing to do with P-P. The usual SS ones simply repeat a current input with neg. unity gain (normally).
(As an aside, it is also possible to design P-P so it does not cancel even harmonics, by using true complementary current rather than the conventional complementary voltage drive, ie, it will sound like SE. This was brought up in another thread. SE from P-P )
The tube versions I spoke of use tubes, and have current gain also, so act as general purpose gain stages, not much different from a conventional grounded cathode stage, except they provide current gain. One can build a SE amplifier using VT current mirrors. (but will require NFB to lower the output Z) They are not nearly as easy to use as conventional gain stages however. (at least until someone makes some custom VT diodes anyway)
Thus far their main attraction is that they can provide distortion free gain (in theory anyway) into a real load impedance, which other gain stages cannot do. Well, there is also the voltage mirror, mentioned in the same thread, which can (again in theory) provide distortion free gain (but not into a real resistive load impedance). The first stage of the Aikido pre-amp uses essentially the voltage mirror configuration for distortion cancelllation by the way.
EC has been applied to SS current mirrors successfully, so it should be possible for tube based ones also, but tube count goes up 2X.
Don
I would try any experiments with screen drive out on something cheaper first, before using the 4X150s. Gain some experience first. ( perhaps I should warn you that I ONLY experiment with new ideas, untread territory for tubes, I'm BORED with the usual stuff, so don't expect canned, proven solutions here.)
The Berning-SwitchMode OT should work for EC designs, however a few caveats apply. It requires a low pass filter to remove switching artifacts, so bandwidth is limited (as conventionally constructed anyway, I have a 4 phase version that does not have these bandwidth limitations due to the switching ) , but usually one would only consider applying EC before the OT anyway.
Next caveat is if you want to do partial CFB with the B-SMOT, the % cathode winding section also gets % fractional "B+" power supply voltage injected, so the cathode is not truly referenced to ground then. This can be made to work, but requires some extra measures. Partial CFB is very useful for pentode outputs and for a unity gain tap point for the EC.
You also have to deal with EMI filter/shielding issues for a conventionally configured B-SMOT. (the 4 phase version however does not generate any significant EMI since it uses controlled switching slew rates.)
Current mirrors conventionally added onto a diffl. amp. stage do indeed become a P-P design. However, current mirrors by themselves have nothing to do with P-P. The usual SS ones simply repeat a current input with neg. unity gain (normally).
(As an aside, it is also possible to design P-P so it does not cancel even harmonics, by using true complementary current rather than the conventional complementary voltage drive, ie, it will sound like SE. This was brought up in another thread. SE from P-P )
The tube versions I spoke of use tubes, and have current gain also, so act as general purpose gain stages, not much different from a conventional grounded cathode stage, except they provide current gain. One can build a SE amplifier using VT current mirrors. (but will require NFB to lower the output Z) They are not nearly as easy to use as conventional gain stages however. (at least until someone makes some custom VT diodes anyway)
Thus far their main attraction is that they can provide distortion free gain (in theory anyway) into a real load impedance, which other gain stages cannot do. Well, there is also the voltage mirror, mentioned in the same thread, which can (again in theory) provide distortion free gain (but not into a real resistive load impedance). The first stage of the Aikido pre-amp uses essentially the voltage mirror configuration for distortion cancelllation by the way.
EC has been applied to SS current mirrors successfully, so it should be possible for tube based ones also, but tube count goes up 2X.
Don
I see. But the Aleph-X is already differential output, so that kind of summing wouldn't make sense. (BTW, R46 and R47 just look like current feedback, not as in sampling output current but in the sense they use in current-feedback opamps, with feedback going into a low impedance input; I think people put them in to reduce DC offset, they weren't in the original Aleph-X schematic).
Can you post more info on your implementation of the switchmode OT? I got an assortment of large pot-core ferrites from surplussales and eBay, and was thinking of doing an implementation once I finished some other things.
I won't be experimenting with the screen drive for some time. I managed to break the sapphire layer of my microhollow cathodes by clamping too hard, and I have to machine new ones, so I'm putting this off for a while...
Can you post more info on your implementation of the switchmode OT? I got an assortment of large pot-core ferrites from surplussales and eBay, and was thinking of doing an implementation once I finished some other things.
I won't be experimenting with the screen drive for some time. I managed to break the sapphire layer of my microhollow cathodes by clamping too hard, and I have to machine new ones, so I'm putting this off for a while...
The 4 phase Switch Mode OT (4-SMOT) is simple conceptually.
Take two of the B-SMOT circuits and connect together at the load and at the tube(s). Same tubes and same + and - LV power supplies still. Then the 180 degree square wave drive signals to the MosFets get shortened to 100+ degree pulses with slow edges (can use up to 40 degrees for turn-on slewing and 40 deg. for turn-off slewing, with solid on during middle 100+ deg.).
The drives for one unit are 90 degrees shifted with respect to the other unit. With the 4 phases, the tubes are always 100% connected to the load. No momentary glitches of disconnect at the edges. (the diode bridges do the commutating between overlapping phases for the tubes smoothly if diodes are matched) The slow turn-on and turn-off prevent EMI generation. The slow turn-on and turn-off do not cause any efficency loss either since the current always flows thru the lowest resistance path which is always one solidly on MosFet(s). No low pass filter needed. Very high bandwidth.
Obviously this won't be catching on commercially any time soon.
Some simplifications are possible. The Mosfet quantity can be cut in half by using center tapped xfmr windings for the MosFet switches instead of full switching bridges.
One can reduce down to a single LV power supply with further complexity by using two auxiliary switching inverters to generate pulsed (100+ deg.) synchronous AC power (90 deg phased between them). The sync. AC power gets inserted in series with the OT xfmers sec. or pri. windings. The old + and - LV power supply points getting grounded then. These aux. inverters can then generate the other HV B+s needed for the rest of the amplifier stages as well, so is not as bad as it seems.
If one can find some MosFets/switches that do not have parasitic reverse diodes and are bidirectional current conducting, one can reduce this even further down to just one ferrite OT xfmr and two switches for it, surprisingly. The two 90 deg. phased aux. sync. supplies just have to connect to two the tube side windings on the same xfmr before the bridge rectifiers. A bit complex to explain, but is using synchronous detection with opposite phases on the two P-P tubes. I need to check IGBTs to see if they will do this, probably not. But might be able to resort to a back to back sampling schottky diode bridge driven by one aux. inverter.
Trading parts for design complexities. Then only 2 MosFets used for each aux. inverter. So half the parts of the original B-SMOT and still using 4 phases, only one LV power supply, and B+s for the rest of the amp for free! With high bandwidth and low EMI.
Can simplify even further for a SE design.
Don
Take two of the B-SMOT circuits and connect together at the load and at the tube(s). Same tubes and same + and - LV power supplies still. Then the 180 degree square wave drive signals to the MosFets get shortened to 100+ degree pulses with slow edges (can use up to 40 degrees for turn-on slewing and 40 deg. for turn-off slewing, with solid on during middle 100+ deg.).
The drives for one unit are 90 degrees shifted with respect to the other unit. With the 4 phases, the tubes are always 100% connected to the load. No momentary glitches of disconnect at the edges. (the diode bridges do the commutating between overlapping phases for the tubes smoothly if diodes are matched) The slow turn-on and turn-off prevent EMI generation. The slow turn-on and turn-off do not cause any efficency loss either since the current always flows thru the lowest resistance path which is always one solidly on MosFet(s). No low pass filter needed. Very high bandwidth.
Obviously this won't be catching on commercially any time soon.
Some simplifications are possible. The Mosfet quantity can be cut in half by using center tapped xfmr windings for the MosFet switches instead of full switching bridges.
One can reduce down to a single LV power supply with further complexity by using two auxiliary switching inverters to generate pulsed (100+ deg.) synchronous AC power (90 deg phased between them). The sync. AC power gets inserted in series with the OT xfmers sec. or pri. windings. The old + and - LV power supply points getting grounded then. These aux. inverters can then generate the other HV B+s needed for the rest of the amplifier stages as well, so is not as bad as it seems.
If one can find some MosFets/switches that do not have parasitic reverse diodes and are bidirectional current conducting, one can reduce this even further down to just one ferrite OT xfmr and two switches for it, surprisingly. The two 90 deg. phased aux. sync. supplies just have to connect to two the tube side windings on the same xfmr before the bridge rectifiers. A bit complex to explain, but is using synchronous detection with opposite phases on the two P-P tubes. I need to check IGBTs to see if they will do this, probably not. But might be able to resort to a back to back sampling schottky diode bridge driven by one aux. inverter.
Trading parts for design complexities. Then only 2 MosFets used for each aux. inverter. So half the parts of the original B-SMOT and still using 4 phases, only one LV power supply, and B+s for the rest of the amp for free! With high bandwidth and low EMI.
Can simplify even further for a SE design.
Don
I forgot to mention that all MosFets/switches (above scheme) are operating on the Low Voltage (LV) output side of the OT xfmr(s). I would probably use IRF540 Mosfets with their .008 Ohm ON resistance. The only HV parts are the bridge rectifiers for the tubes and the tubes themselves. Even the final sync. detection scheme uses no HV switches.
Some finer points such as leakage inductance currents in the xfmrs must be dealt with. The newer MOSFETs have internal diodes that can handle these (gets dumped back into the LV power supply(s)) otherwise have to add prot. diodes.
Don
Some finer points such as leakage inductance currents in the xfmrs must be dealt with. The newer MOSFETs have internal diodes that can handle these (gets dumped back into the LV power supply(s)) otherwise have to add prot. diodes.
Don
I just asked in a thread on the high-powered version, and the reply was "Less than 0.1% at rated power, and less than 0.01% through most of the powerband." The profile is probably similar to that in the figure in http://www.passlabs.com/downloads/articles/susy.pdf (p.1 bottom right) although that amp is one cascoded stage instead of two stages, and doesn't use a modulated current source as the Aleph-X does.smoking-amp said:
Yes, just use two SE schematics for making a 4 phase SE, or two P-P schematics for a 4 phase P-P. Same simplifications apply from there.
Edge slopes can just use an RC filter on the MosFet gates to slow them down, double RC can give nice S shaped edges. Linear Tech. has a chip that goes even further, it controls the slew rate of both voltage and current of MosFets, but is probably overkill for this. The slow slewing means we don't even need a fancy MosFet driver chip, just RC the outputs of the TTL or CMOS digital ICs generating the control signals. El-Cheapo should work fine here. My 4 phase controller cost about $2.50 (5 CMOS chips).
The amplitude modulation scheme, as usually presented, requires HV switching transistors to modulate the tube output signal. This causes reliability problems. My approach using the sychronous auxiliary inverter supplies gets around this problem by generating pulsed HV power to begin with (straight from the aux. ferrite xfmr secondary(s)), otherwise is a very similar scheme. The tube(s) of course still sees nice smooth HV DC power since the bridge rectifiers for each phase combine to give solid DC at the tube(s). But the AC phase currents get modulated by the tube(s) and passed thru the ferrite OT xfmr(s) and demodulated by the switches on the LV output side. I just did the earlier explanation by adapting step-wise from the Berning approach, but it is really just a more clever version of the synchronous modulation/demodulation scheme. They are all very similar actually.
Another approach, not explored yet, is to use a PWM converter like the Cuk converter design. This can give zero ripple output even with varying pulse width modulation. This would allow one to vary the turns ratio of the OT by just adjusting the pulse duty cycle. (A knob on the controller to set primary Z!) A simple PWM chip would suffice for control. Uncertainty here is the varying load impedance of the usual speakers. Ripple free PWM converters, like the Cuk, prefer constant load impedance. Of course, if one had a microP controller chip instead of a simple PWM chip controller, could probably correct for this Z load variance real time. This would require a hefty development effort though. Maybe could get away with just a current sensing PWM chip, don't know.
Don
Edge slopes can just use an RC filter on the MosFet gates to slow them down, double RC can give nice S shaped edges. Linear Tech. has a chip that goes even further, it controls the slew rate of both voltage and current of MosFets, but is probably overkill for this. The slow slewing means we don't even need a fancy MosFet driver chip, just RC the outputs of the TTL or CMOS digital ICs generating the control signals. El-Cheapo should work fine here. My 4 phase controller cost about $2.50 (5 CMOS chips).
The amplitude modulation scheme, as usually presented, requires HV switching transistors to modulate the tube output signal. This causes reliability problems. My approach using the sychronous auxiliary inverter supplies gets around this problem by generating pulsed HV power to begin with (straight from the aux. ferrite xfmr secondary(s)), otherwise is a very similar scheme. The tube(s) of course still sees nice smooth HV DC power since the bridge rectifiers for each phase combine to give solid DC at the tube(s). But the AC phase currents get modulated by the tube(s) and passed thru the ferrite OT xfmr(s) and demodulated by the switches on the LV output side. I just did the earlier explanation by adapting step-wise from the Berning approach, but it is really just a more clever version of the synchronous modulation/demodulation scheme. They are all very similar actually.
Another approach, not explored yet, is to use a PWM converter like the Cuk converter design. This can give zero ripple output even with varying pulse width modulation. This would allow one to vary the turns ratio of the OT by just adjusting the pulse duty cycle. (A knob on the controller to set primary Z!) A simple PWM chip would suffice for control. Uncertainty here is the varying load impedance of the usual speakers. Ripple free PWM converters, like the Cuk, prefer constant load impedance. Of course, if one had a microP controller chip instead of a simple PWM chip controller, could probably correct for this Z load variance real time. This would require a hefty development effort though. Maybe could get away with just a current sensing PWM chip, don't know.
Don
OH, OK.
Well, if one inserts the carrier at the plate voltage of the output tube one has almost the same scheme as mine (the overlapping phases approach doesn't need low pass filtering though). Putting the carrier in earlier will require more tube stages to handle the high frequency, but should work fine. Either way, the secondary side of the ferrite xfmr needs to demodulate the signal back to the audio realm.
To get + and - audio signal out one still needs sychronous switch demodulation as in the Berning or 4 phase approaches. (too opposed diode bridges don't work, they just short each other out) I suppose one could use a SE scheme with just a single polarity of rectification output (two diodes, CT secondary) with a current source to opposite polarity (not too efficient though) or, Hmmm....., maybe could could just use an inductor from output back to ground (instead of the current source). Seems like that should work, but now one is back to a great big expensive air gapped inductor, sort of defeats the purpose.
In any case, one needs a low pass filter to remove the carrier from the audio, this will hurt the output Z for damping some, but global NFB could rescue.
Don
getting late here, signing off
Well, if one inserts the carrier at the plate voltage of the output tube one has almost the same scheme as mine (the overlapping phases approach doesn't need low pass filtering though). Putting the carrier in earlier will require more tube stages to handle the high frequency, but should work fine. Either way, the secondary side of the ferrite xfmr needs to demodulate the signal back to the audio realm.
To get + and - audio signal out one still needs sychronous switch demodulation as in the Berning or 4 phase approaches. (too opposed diode bridges don't work, they just short each other out) I suppose one could use a SE scheme with just a single polarity of rectification output (two diodes, CT secondary) with a current source to opposite polarity (not too efficient though) or, Hmmm....., maybe could could just use an inductor from output back to ground (instead of the current source). Seems like that should work, but now one is back to a great big expensive air gapped inductor, sort of defeats the purpose.
In any case, one needs a low pass filter to remove the carrier from the audio, this will hurt the output Z for damping some, but global NFB could rescue.
Don
getting late here, signing off
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.