I think the situation is more complex. I try to write what I have in mind. Amplifier is class (A)B, not classA.
1. The high order harmonics are the most "not welcomed" in power amp audio reproduction. I've experimented with this, and found it is true. Low order harmonic, no matter odd or even (2nd or 3rd) is more tolerable to ear.
2. High harmonics happened in output transistor (in amp with NFB) when there is a sudden change of position, like off--suddenly--on. It creates high order artifacts. But if we make it never turn off or gaining voltage not from dead zero, the sudden burst disappears. Imagine a little fire spark when a relay closed. That is burst of high order distortions.
3. The on-off schema in output stage is quite complex. Imagine you have Complementary EF with RE of 0.1ohm. VBE at 100mA=0.6V. If you have complementary pair, then at no signal (bias 100mA) you will set the VBE multiplier for 0.6V+100mAx0.1ohm+100mAx0.1ohm+0.6V=1.22V. This VBE multiplier will ALWAYS have this exact drop, 1V22 at any condition.
4. Now you got a situation where the amp has to give positive output voltage, has to deliver about 1A positively to the speaker.
In this condition, VBE of upper transistor is not 0.6V anymore, lets say at 1A the VBE is 0.8V.
Then if we wanted to have the lower transistor also on at 100mA (to maintain non-switchoff condition for all output transistors), then the bias have to be:
0.8V+1Ax0.1ohm+100mAx0.1ohm+0.6V=1V51.
You can see, that with settled VBE of 1V22, while the needed VBE to maintain all transistor on=1V51. ---- This means with 1V51 needed, there is only 1V22 supplied, the lower transistor must be OFF. This is simple example. With darlington or drivers, the difference is bigger.
Ordinary VBE multiplier cannot maintain VBE drop needed to maintain all transistor on at all conditions. With ordinary VBE multiplier, high order burst will always happened because the output transistor will on-off-on-off all the time. (remember point 1 and 2, why we do not want on-off-on-off).
This is worse if we use feedback. Gives higher order distortion, due to feedback.
Smart thinking about how to overcome this has been done by NP (with smart bias), Steven, and a Japanese guy with a patent. But unfortunately, they all use quite heavily biased output stage, so the voltage drops in resistors are quite big to be useful.
Since for class B amp, the voltage drops are so little, I wanted to find a way to find a way to "always have all output transistors never turn off".
I have an idea that approaches this not by manipulating VBE multiplier (it can give "positive feedback", can yield in another high order distoriton), but rather pushing the non-off condition by outside force (by using ccs or passive way) that do not affect the signals at all.
Someone has idea on this? Putting CCS with diode to VBE multiplier, while from VBE multiplier there is R to output maybe working.
1. The high order harmonics are the most "not welcomed" in power amp audio reproduction. I've experimented with this, and found it is true. Low order harmonic, no matter odd or even (2nd or 3rd) is more tolerable to ear.
2. High harmonics happened in output transistor (in amp with NFB) when there is a sudden change of position, like off--suddenly--on. It creates high order artifacts. But if we make it never turn off or gaining voltage not from dead zero, the sudden burst disappears. Imagine a little fire spark when a relay closed. That is burst of high order distortions.
3. The on-off schema in output stage is quite complex. Imagine you have Complementary EF with RE of 0.1ohm. VBE at 100mA=0.6V. If you have complementary pair, then at no signal (bias 100mA) you will set the VBE multiplier for 0.6V+100mAx0.1ohm+100mAx0.1ohm+0.6V=1.22V. This VBE multiplier will ALWAYS have this exact drop, 1V22 at any condition.
4. Now you got a situation where the amp has to give positive output voltage, has to deliver about 1A positively to the speaker.
In this condition, VBE of upper transistor is not 0.6V anymore, lets say at 1A the VBE is 0.8V.
Then if we wanted to have the lower transistor also on at 100mA (to maintain non-switchoff condition for all output transistors), then the bias have to be:
0.8V+1Ax0.1ohm+100mAx0.1ohm+0.6V=1V51.
You can see, that with settled VBE of 1V22, while the needed VBE to maintain all transistor on=1V51. ---- This means with 1V51 needed, there is only 1V22 supplied, the lower transistor must be OFF. This is simple example. With darlington or drivers, the difference is bigger.
Ordinary VBE multiplier cannot maintain VBE drop needed to maintain all transistor on at all conditions. With ordinary VBE multiplier, high order burst will always happened because the output transistor will on-off-on-off all the time. (remember point 1 and 2, why we do not want on-off-on-off).
This is worse if we use feedback. Gives higher order distortion, due to feedback.
Smart thinking about how to overcome this has been done by NP (with smart bias), Steven, and a Japanese guy with a patent. But unfortunately, they all use quite heavily biased output stage, so the voltage drops in resistors are quite big to be useful.
Since for class B amp, the voltage drops are so little, I wanted to find a way to find a way to "always have all output transistors never turn off".
I have an idea that approaches this not by manipulating VBE multiplier (it can give "positive feedback", can yield in another high order distoriton), but rather pushing the non-off condition by outside force (by using ccs or passive way) that do not affect the signals at all.
Someone has idea on this? Putting CCS with diode to VBE multiplier, while from VBE multiplier there is R to output maybe working.
Hi lumanauw,
Bipolar transistors are carrier/current driven devices. The lowest distortion is obtained when their base drive current is free to flow without regard for base-emitter voltage, this because base-emitter voltage varies with (loudspeaker back emf induced plus delayed NFB loop corrected) signal waveform rise time, not just simplistically applied current. Thus the fastest and most accurate response does not necessarily arise in maintained class-A because the base junctions cannot then as quickly lose all of their previous drive charge, even with 'always on' exponential drive.
Where an output device must provide reasonable audio power it cannot turn off quickly enough if the base-emitter junction is always forward biased into class-A. NFB can cause a push-pull partner to overdrive into cross conduction due to this charge retention, and this can increase high harmonic distortion.
Quad developed their Current Dumper in the 1970s for the very reasons you discuss. This was like class-A corrected class-B, and with modern devices it remains an excellent solution where NFB loop defined higher power output is required. As Eva pointed out in another column, power transistors are capable of very high power audio as long as they are not (load/NFB loop) driven into cross-conduction or oscillation.
http://www.diyaudio.com/forums/showthread.php?s=&postid=525448 post 28
Does your idea cover loudspeaker back emf voltage leading output current ?
Cheers .......... Graham.
Bipolar transistors are carrier/current driven devices. The lowest distortion is obtained when their base drive current is free to flow without regard for base-emitter voltage, this because base-emitter voltage varies with (loudspeaker back emf induced plus delayed NFB loop corrected) signal waveform rise time, not just simplistically applied current. Thus the fastest and most accurate response does not necessarily arise in maintained class-A because the base junctions cannot then as quickly lose all of their previous drive charge, even with 'always on' exponential drive.
Where an output device must provide reasonable audio power it cannot turn off quickly enough if the base-emitter junction is always forward biased into class-A. NFB can cause a push-pull partner to overdrive into cross conduction due to this charge retention, and this can increase high harmonic distortion.
Quad developed their Current Dumper in the 1970s for the very reasons you discuss. This was like class-A corrected class-B, and with modern devices it remains an excellent solution where NFB loop defined higher power output is required. As Eva pointed out in another column, power transistors are capable of very high power audio as long as they are not (load/NFB loop) driven into cross-conduction or oscillation.
http://www.diyaudio.com/forums/showthread.php?s=&postid=525448 post 28
Does your idea cover loudspeaker back emf voltage leading output current ?
Cheers .......... Graham.
Hi, Graham Maynard,
I'm a DIYer anyway.
In the other thread where they discuss about including or not including output stage to NFB loop, there is opinion that not including output stage to closed loop sounds better.
I think this also have relation with output stage distortion (this very crossover distortion) besides the back EMF issue you said.
Even if there is no back EMF, the behavior of on-off-on-off abrupts high order artifacts is bad in my opinion. Putting this abrupts into NFB loop makes even higher order distortion, due to "hall of mirror" effect. So it makes sense, not putting output stage into NFB loop sounds better, because there will be less high order distortion.
GM, you makes me confused. Is the transistor needs tobe OFF (or reversed biased to suckout base charge) or maintained ON all the time? Those are 2 contrary opinions.
I found this very interesting, but I cannot figure this out. How can we "force" current base, if Vbe is below 0.6V? I think VBE figure is very tied together to base current. Cannot separate those 2.
Aha, you suggest that output transistor HAVE to be OFF or even reversed biased to suck out remaining charge? I read about this causing crossconduction or "Turn-off" distortion in DougSelf book. This is kind of contrary of what I have in this thread.
Is this what they called "GM Doubling"distortion? Distortion in class AB amp where in near crossover area (where it still plays classA) both transistors are active so the GM are doubled? This GM doubling distortion happens when an output stage in classAB changing from classA to B or classB to A where GM is single or double.
Frankly I havent dig up as far as you did.
I'm a DIYer anyway.In the other thread where they discuss about including or not including output stage to NFB loop, there is opinion that not including output stage to closed loop sounds better.
I think this also have relation with output stage distortion (this very crossover distortion) besides the back EMF issue you said.
Even if there is no back EMF, the behavior of on-off-on-off abrupts high order artifacts is bad in my opinion. Putting this abrupts into NFB loop makes even higher order distortion, due to "hall of mirror" effect. So it makes sense, not putting output stage into NFB loop sounds better, because there will be less high order distortion.
GM, you makes me confused.
Is the transistor needs tobe OFF (or reversed biased to suckout base charge) or maintained ON all the time? Those are 2 contrary opinions.
Hi,
speaking about transconductance doubbling, Leach says that this problem does not exsist if i now understood his PDF article.
Cheers
speaking about transconductance doubbling, Leach says that this problem does not exsist if i now understood his PDF article.
Cheers
Regardless of whether or not transconductance doubling exists, it is true that biasing the output stage such that there is a region where both transistors are conducting will add distortion. The optimum is to bias them in class B just below class AB, so that the switching from one to the other occurs simultaneously.
But still, no amount of fancy biasing can ever completely remove crossover distortion - it takes some form of error correction to do that.
P.S. Ultima Thule, "xstr" is an abbreviation for transistor. The letter x is often substituted for "trans" is many words, such as xformer for transformer, or xmit for transmit.
But still, no amount of fancy biasing can ever completely remove crossover distortion - it takes some form of error correction to do that.
P.S. Ultima Thule, "xstr" is an abbreviation for transistor. The letter x is often substituted for "trans" is many words, such as xformer for transformer, or xmit for transmit.
One of ENZO's comments, above, is worth focusing on - crossover distortion is one thing switch-on/off distortion is another. They are obviously closely associated but the mechanisms are different. It is helpful when discussing one or both to be clear which one is being considered. Looking back through the preceding thread, it is not always clear (to me anyway) which is being discussed at any particular point.
------------------------
Mr. Evil wrote
At least in the case of EF output sections this can be attacked by connecting the emitters of the drivers to each other via a resistor (100-300R) rather than to the output. When the opposite device turns on it forces the other off. This is probably not a perfect solution but it seems to work better than just tying them to the output along with the O/P devices.
------------------------
Mr. Evil wrote
At least in the case of EF output sections this can be attacked by connecting the emitters of the drivers to each other via a resistor (100-300R) rather than to the output. When the opposite device turns on it forces the other off. This is probably not a perfect solution but it seems to work better than just tying them to the output along with the O/P devices.
Hi lumanauw,
Is it the opinions that are contrary, or the different but inter-related aspects of amplifier design problems ?
An output device might be statically biased for say 35mA quiescent current, and the base-emitter voltage might be measured at 0.6V, but that will not necessarily be its voltage for every other instant of 35mA conduction during high audio frequency drive conditions. Even emitter follower output stages do not directly follow VAS/CCS potential due to a varying load induced voltage drop that is not directly related to signal input.
I do not suggest that an output transistor HAVE to be OFF etc., but I do note limitations related to active operation.
Cross conduction is due to not instantaneous switch-off (which may be countered as sam9 suggests) but this is not the same as when both devices conduct with a pre-set fixed bias current.
GM doubling means lower distortion when both devices are linearly active. What happens is that the eventual crossover into class-B operation distortion is voltage shifted away from the zero voltage crossover axis, and this is much more likely to become audible when a complex dynamic loudspeaker is the load.
I would not choose to use non-class-A operating output devices that are not within a closed NFB loop, in case loudspeaker induced class-AB crossover distortion might arise.
Cheers .......... Graham.
Is it the opinions that are contrary, or the different but inter-related aspects of amplifier design problems ?
An output device might be statically biased for say 35mA quiescent current, and the base-emitter voltage might be measured at 0.6V, but that will not necessarily be its voltage for every other instant of 35mA conduction during high audio frequency drive conditions. Even emitter follower output stages do not directly follow VAS/CCS potential due to a varying load induced voltage drop that is not directly related to signal input.
I do not suggest that an output transistor HAVE to be OFF etc., but I do note limitations related to active operation.
Cross conduction is due to not instantaneous switch-off (which may be countered as sam9 suggests) but this is not the same as when both devices conduct with a pre-set fixed bias current.
GM doubling means lower distortion when both devices are linearly active. What happens is that the eventual crossover into class-B operation distortion is voltage shifted away from the zero voltage crossover axis, and this is much more likely to become audible when a complex dynamic loudspeaker is the load.
I would not choose to use non-class-A operating output devices that are not within a closed NFB loop, in case loudspeaker induced class-AB crossover distortion might arise.
Cheers .......... Graham.
Hi, Everybody,
In amplifier design, there are 3 stages, differential, VAS, output stage. In differential and VAS, there is much less problem, because they all operate in classA or even single ended.
That leave us with output stage (which has the biggest distortion from 3 stages, I'm talking about class AB amp).
People usually concentrate on differential and VAS design and less think about classAB output stage. Just take it and accept it as it is for now this long years. Maybe there is not much to do here? Breakthroughs like NP's smart bias, Hawksford and NP-PMA Error Correction are very rare.
1. LSN distortion (Large Signal Nonlinearities). This happens with BJT output stage, that have to swing big voltage and big current. When these 2 big swings occur, the BJT cannot maintain HFE figure--it will fluctuate and making distortion. Class A still have this, but less, since the current fluctuation is less.
2. Crossover distortion : happens in class AB output stage, where the NPN or PNP are one=on and the other=off. Why it is one on and one off? Because all this time we use VBE multiplier, it results the mechanism like post#21.
This happens because bipolar transistor are not linear device from 0V. It takes 0-0.6V dead zone, after 0.6V it begans active. If we have bipolar that is on linearly from 0V, we wouldn't have crossover distortion, since there is no abrupt position change at 0.6V (from off suddenly on with big gain). This abrupt position change is the one who is responsible for generating high order distortions in classAB output stage.
3. Turn-off distortion. This happens when a bipolar transistor is not OFF when it should have been OFF, because there is still charges at the base. This charge have tobe removed or reversed biased to suck out charges. The output type that can do this is EF type II in DougSelf book, like Sam9 suggested below :
Other kind of Output stage, like CFP bipolars, cannot discharge this, because it has no path to discharge. To reverse bias in CFP we will need a voltage point bigger than rail voltage. It can be make less with smaller CFP resistor, but not 100%.
In CFP, this turnoff distortion can be dangerous, it wil makes crossconduction from NPN-PNP transistor in higher frequencies.
So, it is right, Crossover distortion and Turn-off distortion are different distortion mechanism.
Having output transistor that never turn-off will cope with no.2 and no.3 mechanism simultaneously.
Yep, thats what I intended. How to do never-off output stage in classAB amp. Every possibility that I have right now seems suggests to leave ordinary VBE multiplier or having outside passively CCS controlling output transistor not turning off. But haven't come out with one yet that works.
In amplifier design, there are 3 stages, differential, VAS, output stage. In differential and VAS, there is much less problem, because they all operate in classA or even single ended.
That leave us with output stage (which has the biggest distortion from 3 stages, I'm talking about class AB amp).
People usually concentrate on differential and VAS design and less think about classAB output stage. Just take it and accept it as it is for now this long years. Maybe there is not much to do here? Breakthroughs like NP's smart bias, Hawksford and NP-PMA Error Correction are very rare.
Yes, classA output stage is better than AB. Can we make classAB better? Making it having no (less) high order distortion is good enough for me.
???????????????????????
Yes, they are different. From what I read, classAB major drawbacks are 3:
1. LSN distortion (Large Signal Nonlinearities). This happens with BJT output stage, that have to swing big voltage and big current. When these 2 big swings occur, the BJT cannot maintain HFE figure--it will fluctuate and making distortion. Class A still have this, but less, since the current fluctuation is less.
2. Crossover distortion : happens in class AB output stage, where the NPN or PNP are one=on and the other=off. Why it is one on and one off? Because all this time we use VBE multiplier, it results the mechanism like post#21.
This happens because bipolar transistor are not linear device from 0V. It takes 0-0.6V dead zone, after 0.6V it begans active. If we have bipolar that is on linearly from 0V, we wouldn't have crossover distortion, since there is no abrupt position change at 0.6V (from off suddenly on with big gain). This abrupt position change is the one who is responsible for generating high order distortions in classAB output stage.
3. Turn-off distortion. This happens when a bipolar transistor is not OFF when it should have been OFF, because there is still charges at the base. This charge have tobe removed or reversed biased to suck out charges. The output type that can do this is EF type II in DougSelf book, like Sam9 suggested below :
Other kind of Output stage, like CFP bipolars, cannot discharge this, because it has no path to discharge. To reverse bias in CFP we will need a voltage point bigger than rail voltage. It can be make less with smaller CFP resistor, but not 100%.
In CFP, this turnoff distortion can be dangerous, it wil makes crossconduction from NPN-PNP transistor in higher frequencies.
So, it is right, Crossover distortion and Turn-off distortion are different distortion mechanism.
Having output transistor that never turn-off will cope with no.2 and no.3 mechanism simultaneously.
Yep, thats what I intended. How to do never-off output stage in classAB amp. Every possibility that I have right now seems suggests to leave ordinary VBE multiplier or having outside passively CCS controlling output transistor not turning off. But haven't come out with one yet that works.
Hi all,
I'm thinking it's not just the turning off of the transistors in the out stage but the handing over (pnp npn) that is the problem. If you were to have them both equaly on through the whole swing tossing the load back and forth you'd have class A and there'd be no xover-d. Aything else you will have xover-d to some degree regardless of your bias. If your bias is above your amplifier's peak level then you effectively have class A. If not, when one transistor lets go there will be x-over-d, Al-beit though not really audible at high bias levels.
I'm thinking it's not just the turning off of the transistors in the out stage but the handing over (pnp npn) that is the problem. If you were to have them both equaly on through the whole swing tossing the load back and forth you'd have class A and there'd be no xover-d. Aything else you will have xover-d to some degree regardless of your bias. If your bias is above your amplifier's peak level then you effectively have class A. If not, when one transistor lets go there will be x-over-d, Al-beit though not really audible at high bias levels.
smoothing the crossover
The effective output stage transconductance doesn't stay constant during the crossover as Self shows in his book or Leach in his paper. The wrinkle in the transconductance around crossover needs to be smoothed out. Tube amps with their 3/2 power transfer law seem to do a better job of this. So here is an idea to try (simulator first), put a high current thermionic diode (like a damper tube rectifier) in series (correct polarity for conduction) with the base terminal of each emitter follower output. Now the voltage to current transfer will be much smoother. One will have to boost the VBE bias voltage considerably to get class AB conduction going. Just an idea. Tube sound maybe?
Don
The effective output stage transconductance doesn't stay constant during the crossover as Self shows in his book or Leach in his paper. The wrinkle in the transconductance around crossover needs to be smoothed out. Tube amps with their 3/2 power transfer law seem to do a better job of this. So here is an idea to try (simulator first), put a high current thermionic diode (like a damper tube rectifier) in series (correct polarity for conduction) with the base terminal of each emitter follower output. Now the voltage to current transfer will be much smoother. One will have to boost the VBE bias voltage considerably to get class AB conduction going. Just an idea. Tube sound maybe?
Don
Hi lumanauw,
You are covering the situation exactly as Douglas Self did, but he completely ignored the effect of amplifier propagation delay and thus the output stage commutation error due to reverse driven output device crossover when resultant loudspeaker reactivity shifts the zero current crossover point away from the zero ac volt line. This is why we hear distortion that does not show up with resistor testing.
When this happens it is the loudspeaker that is controlling the amplifier's output requirement wrt input, but before the output stage gets the correct NFB loop controlled drive it will have already fractionally crossed over output device conduction, which the differential plus VAS stages must correct, and which might be slowed by their Miller connected VAS C.dom.
With straightforward circuitry and resistor load observation, crossover distortion itself can be reduced to phenomenally low levels in class-A, but a crossover distortion like (reverse commutation) effect can still arise when the load is a loudspeaker. Did not Self himself argue that crossover distortion could be rendered negligible ?
It is the very fact that a loudspeaker can cause an offset shift between class-A and class-B operation in a class-AB output stage that leaves a NFB loop controlled amplifier susceptible to produce conduction crossover spikes, and no amount of NFB can prevent this because the distortion arises as quiescent bias controlled flow is overcome.
Hence progression to the Zen type of amplifier.
I am puzzled how you expect your CCS to not affect pre-set Vbe bias. If neither output devices loses conduction then it is NOT class-AB, and this is where 'exponential' drive types can retain control.
However, 'exponential' drive, as once illustrated I think by Steven, might not have good gain at low amplitude levels and will be slower falling out of power conduction, so you could still end up with more distortion than is possible with a carefully designed class-AB.
The Quad Current Dumper uses a fractionally delaying output bridge to class-A correct class-B error, and it does work, though the early ones were not as perfect as originally suggested.
Cheers .......... Graham.
You are covering the situation exactly as Douglas Self did, but he completely ignored the effect of amplifier propagation delay and thus the output stage commutation error due to reverse driven output device crossover when resultant loudspeaker reactivity shifts the zero current crossover point away from the zero ac volt line. This is why we hear distortion that does not show up with resistor testing.
When this happens it is the loudspeaker that is controlling the amplifier's output requirement wrt input, but before the output stage gets the correct NFB loop controlled drive it will have already fractionally crossed over output device conduction, which the differential plus VAS stages must correct, and which might be slowed by their Miller connected VAS C.dom.
With straightforward circuitry and resistor load observation, crossover distortion itself can be reduced to phenomenally low levels in class-A, but a crossover distortion like (reverse commutation) effect can still arise when the load is a loudspeaker. Did not Self himself argue that crossover distortion could be rendered negligible ?
It is the very fact that a loudspeaker can cause an offset shift between class-A and class-B operation in a class-AB output stage that leaves a NFB loop controlled amplifier susceptible to produce conduction crossover spikes, and no amount of NFB can prevent this because the distortion arises as quiescent bias controlled flow is overcome.
Hence progression to the Zen type of amplifier.
I am puzzled how you expect your CCS to not affect pre-set Vbe bias. If neither output devices loses conduction then it is NOT class-AB, and this is where 'exponential' drive types can retain control.
However, 'exponential' drive, as once illustrated I think by Steven, might not have good gain at low amplitude levels and will be slower falling out of power conduction, so you could still end up with more distortion than is possible with a carefully designed class-AB.
The Quad Current Dumper uses a fractionally delaying output bridge to class-A correct class-B error, and it does work, though the early ones were not as perfect as originally suggested.
Cheers .......... Graham.
Hi lumanauw,
I have been separately thinking about your overall aims.
You want more power than class-A could realistically provide.
You want low crossover/output stage distortion.
Why not try a single Darlington driver to multi-paralleled output devices, maybe even a x4 or x5 bank of lower 'C' medium power driver devices.
Darlington will ensure driving capability, multiple output devices conducting less 'hard' will not be so slow at turning off.
If you go for class-AB with say a shared 1A quiescent current, even the most reactive loudspeaker is unlikely to reverse induce output device bias commutation at normal listening levels.
Distortion could be very low as all devices operate at near optimum gain/bandwidth.
Cheers .......... Graham.
I have been separately thinking about your overall aims.
You want more power than class-A could realistically provide.
You want low crossover/output stage distortion.
Why not try a single Darlington driver to multi-paralleled output devices, maybe even a x4 or x5 bank of lower 'C' medium power driver devices.
Darlington will ensure driving capability, multiple output devices conducting less 'hard' will not be so slow at turning off.
If you go for class-AB with say a shared 1A quiescent current, even the most reactive loudspeaker is unlikely to reverse induce output device bias commutation at normal listening levels.
Distortion could be very low as all devices operate at near optimum gain/bandwidth.
Cheers .......... Graham.
Hi lumenauw,
This is the circuit I came up with in yr2000. It is half of an output stage.
Class-AB operation is set by the 100 ohm resistor. For constant class-A with an approximate exponential characteristic omit the 100 ohm resistor.
This arrangement has a sliding gain characteristic;-
At low current the output transistors control gain plus stability;
then the BC337s (or a single slightly higher current small transistor) provide additional low gain by denying full bias drive to the driver transistor;
for high power output the BC337s are fully conducting with all drive increase operating the driver.
When applied to a JLH amplifier circuit this sub circuit behaves just like the original with say 0.5 to 1A quiescent current, and it is the output devices alone that remain normally active, but either half can be driven into very high power drive as demanded by the loudspeaker loading. This is Mosfet like but without the oscillation risks, and load invariant like but without a sharp crossover point.
Distortion with class-AB operation is lower at higher power than with pure class-A operation, yet still without crossover distortion as the conducting half retains very gradual control. Distortion with full class-A at lower powers can extremely low.
This is one to think about, but there's always more than one way to achieve results. All designs are contrary, each having good and bad points.
Cheers .............. Graham
This is the circuit I came up with in yr2000. It is half of an output stage.
Class-AB operation is set by the 100 ohm resistor. For constant class-A with an approximate exponential characteristic omit the 100 ohm resistor.
This arrangement has a sliding gain characteristic;-
At low current the output transistors control gain plus stability;
then the BC337s (or a single slightly higher current small transistor) provide additional low gain by denying full bias drive to the driver transistor;
for high power output the BC337s are fully conducting with all drive increase operating the driver.
When applied to a JLH amplifier circuit this sub circuit behaves just like the original with say 0.5 to 1A quiescent current, and it is the output devices alone that remain normally active, but either half can be driven into very high power drive as demanded by the loudspeaker loading. This is Mosfet like but without the oscillation risks, and load invariant like but without a sharp crossover point.
Distortion with class-AB operation is lower at higher power than with pure class-A operation, yet still without crossover distortion as the conducting half retains very gradual control. Distortion with full class-A at lower powers can extremely low.
This is one to think about, but there's always more than one way to achieve results. All designs are contrary, each having good and bad points.
Cheers .............. Graham
Attachments
Hi, Graham Maynard,
I would need re-read to understand throughly your explenation.
It seems that you say that speaker induced (back EMF) is more important than xover distortion.
I'm confused now with R load only. What a confusion to get it right for real speaker load.
Putting only R drop for preset bias. But this also don't have temperature compensation properties. I think about putting R and diode in series, where the diode is put to heatsink to give temperature compensation. If people look at it, it would be a "stone age bias setup"
I would need re-read to understand throughly your explenation. It seems that you say that speaker induced (back EMF) is more important than xover distortion.
You are right. All my assumptions are based if the load is purely R. I haven't think about how the situation is if faced with real speaker. At least you have pointed me the direction.
I'm confused now with R load only. What a confusion to get it right for real speaker load.
As for now, I think one way to have preset bias that follows the need to make all active device is simple R. VBE multiplier cannot be used for this issue, because the bias drop is always fixed value. I need a variabel bias value for top and bottom output device that always maintain minimal bias (all transistor never OFF). That will be an answer from stone age wouldn't it?
Putting only R drop for preset bias. But this also don't have temperature compensation properties. I think about putting R and diode in series, where the diode is put to heatsink to give temperature compensation. If people look at it, it would be a "stone age bias setup"Yeah, that is what I aimed. What should it be called?
I try to advoid complex cct like that (or EC). In a certain condition, it gives gain/positive feedback, this will result in another high order distortion, if it is included in the whole amp closed loop. Try to shoot the problem with as simple CCT as possible.
Another requirement : the amp should run cool, not hot
I still focus on way to make suitable bias method (not using ordinary VBE multiplier). I will be able to use different kind of output configuration if I have know the answer to this.
I don't understand this. Could you explain a little more?
Hi darkfenriz,
If I understand your idea correctly, you are trying to get a non-constant voltage between the upper and lower output transistors to perform sliding bias by means of two separate VAS circuits? The required variation of bias for sliding bias is fairly small, I think, so it would be difficult to get sufficiently accurate tracking between the two VAS circuits, especially with nonlinearities in each.
Since the sliding bias needs to increase with either polarity of voltage swing, actually output current swing, it would seem that two back to back nonlinearities are called for however in the single VAS output with some sensitivity to output current. Graham's sliding bias works this way. The earlier damper tube bias modifier I mentioned is somewhat similar but puts the nonlinearities in the transfer functions of the two output transistors to try to get a smoothly varying total transconductance when both are active during crossover and at the edges of conduction.
Mosfets as outputs have the interesting property, over part of their transfer curve, of being square law, which means the transconductance (derivative of transfer law) is linearly varying with signal. When P and N channel outputs are put together totem pole wise, the sum of their transconductances is approx. constant. But when the signal finally turns one of them off, there is an abrupt kink in the effective transconductance. Also, Mosfets are more like exponential functions at low currents just like the bipolar transistors, so the same crossover problems crop up near class B operation. By changing the setting of the fixed bias control for Mosfets, to alter the idle current level, one can vary the power law of the Mosfet transfer function somewhat between exponential, to square law, eventually to near linear at high current.
Don
If I understand your idea correctly, you are trying to get a non-constant voltage between the upper and lower output transistors to perform sliding bias by means of two separate VAS circuits? The required variation of bias for sliding bias is fairly small, I think, so it would be difficult to get sufficiently accurate tracking between the two VAS circuits, especially with nonlinearities in each.
Since the sliding bias needs to increase with either polarity of voltage swing, actually output current swing, it would seem that two back to back nonlinearities are called for however in the single VAS output with some sensitivity to output current. Graham's sliding bias works this way. The earlier damper tube bias modifier I mentioned is somewhat similar but puts the nonlinearities in the transfer functions of the two output transistors to try to get a smoothly varying total transconductance when both are active during crossover and at the edges of conduction.
Mosfets as outputs have the interesting property, over part of their transfer curve, of being square law, which means the transconductance (derivative of transfer law) is linearly varying with signal. When P and N channel outputs are put together totem pole wise, the sum of their transconductances is approx. constant. But when the signal finally turns one of them off, there is an abrupt kink in the effective transconductance. Also, Mosfets are more like exponential functions at low currents just like the bipolar transistors, so the same crossover problems crop up near class B operation. By changing the setting of the fixed bias control for Mosfets, to alter the idle current level, one can vary the power law of the Mosfet transfer function somewhat between exponential, to square law, eventually to near linear at high current.
Don
Class H?
Why not just use a class H like amplifier to avoid crossover distortion and get efficiency too? This just uses a small class A amplifier with floating +/- 5 Volt rail voltages manuevered around by a class aB amp. The class A amp portion always has its available rail voltages just above (and below) the signal, and only the class A portion draws constant heavy current from the +/- 5 V supply, so it operates reasonably efficently. This can be added onto any existing conventional class aB amplifier too.
Don
Why not just use a class H like amplifier to avoid crossover distortion and get efficiency too? This just uses a small class A amplifier with floating +/- 5 Volt rail voltages manuevered around by a class aB amp. The class A amp portion always has its available rail voltages just above (and below) the signal, and only the class A portion draws constant heavy current from the +/- 5 V supply, so it operates reasonably efficently. This can be added onto any existing conventional class aB amplifier too.
Don
Hi darkfenriz,
I wonder if you have seen the recent circuit due to Pavel.
He minimises error by providing separate gain tailored driver stages for each output half; maybe someone can point you to his circuit.
Hi Don,
Yes there can be a Mosfet transconductance change at current handover, and this can cause reproduction non-linearity when the loudspeaker momentarily exhibits a characteristic which shifts the device current crossover ahead of the normally set VAS bias voltage crossover, but I have it in the back of my mind that the problem arises due to propagation and NFB loop correction delaying 'within the loop's stabilisation circuitry' which otherwise shows up as generating hf output inductance under resistive test bench examination. Where stabilisation circuitry does not render the amp inductive, and it does not matter what non-class-A circuitry is used, then a leading loudspeaker characteristic is not going to exacerbate any reverse induced crossover (commutation) distortion.
The output stage illustrated above can keep output angle below 10 degrees at 20kHz, whereas so many 'hi-fi' amps are in complete quadrature due to enforced stabilty, as is necessary in class-AB and B with plain Darlington and complementary arrangements, most Mosfets too. I believe Pavel's output circuit could provide such a low angle too, but have not checked.
Do you have any proven class-H we can see ?
Cheers .......... Graham.
I wonder if you have seen the recent circuit due to Pavel.
He minimises error by providing separate gain tailored driver stages for each output half; maybe someone can point you to his circuit.
Hi Don,
Yes there can be a Mosfet transconductance change at current handover, and this can cause reproduction non-linearity when the loudspeaker momentarily exhibits a characteristic which shifts the device current crossover ahead of the normally set VAS bias voltage crossover, but I have it in the back of my mind that the problem arises due to propagation and NFB loop correction delaying 'within the loop's stabilisation circuitry' which otherwise shows up as generating hf output inductance under resistive test bench examination. Where stabilisation circuitry does not render the amp inductive, and it does not matter what non-class-A circuitry is used, then a leading loudspeaker characteristic is not going to exacerbate any reverse induced crossover (commutation) distortion.
The output stage illustrated above can keep output angle below 10 degrees at 20kHz, whereas so many 'hi-fi' amps are in complete quadrature due to enforced stabilty, as is necessary in class-AB and B with plain Darlington and complementary arrangements, most Mosfets too. I believe Pavel's output circuit could provide such a low angle too, but have not checked.
Do you have any proven class-H we can see ?
Cheers .......... Graham.
Hi, Smoking amp,
Some said that mosfets are less suitable for class AB output stages. They are most suitable for full classA output stage.
What makes opinion like that? Is it the xover distortion worse than bipolar or something else making it less suitable for class AB?
http://www.diyaudio.com/forums/showthread.php?s=&threadid=23590&highlight=
But someone said that this amp has problems. I dont know what kind of problem it is, but Technics seems have stopped using this topology
Hi, EVA,
Back to post#3, what is the mechanism about on-off-on-off condition of diodes that you said worse than output stage xover distortion? More abrupt than xover distortion of output stage?
What is the difference in this mechanism between ordinary diode (IN4002), small diode (IN4148), Fast diode (MUR) and schottky?
How is the xover distortion in mosfet output stages? Is it worse than bipolar, or the same, or better?
Some said that mosfets are less suitable for class AB output stages. They are most suitable for full classA output stage.
What makes opinion like that? Is it the xover distortion worse than bipolar or something else making it less suitable for class AB?
This is a CCT used by Technics that similiar to your idea. It is based on Sano Hirota patent. Clever it is.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=23590&highlight=
But someone said that this amp has problems. I dont know what kind of problem it is, but Technics seems have stopped using this topology
Hi, EVA,
Back to post#3, what is the mechanism about on-off-on-off condition of diodes that you said worse than output stage xover distortion? More abrupt than xover distortion of output stage?
What is the difference in this mechanism between ordinary diode (IN4002), small diode (IN4148), Fast diode (MUR) and schottky?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.