Hi,
Thought I'd share some thoughts on class AB.
(I'm making a small 1/4 A AB version of the DXamp.)
No-one has been more responsible than Douglas Self for giving
Class AB a good kicking, so I make no apologies for using
only his examples in my arguments, for reference.
I'll also stick to 1/4 class A, half voltage/current he used.
1) Class AB is worse than Class A and Class B ?
Yes it will sound worse than class A. But worse than class B ?
Just because distortion at near full power is worse than class B ?
Surely that is a non-argument, the pro's and con's not discussed ?
2) 1/4 A AB below -6dB is to all intents and purposes Class A.
Self quotes a typical volume setting of -15dB, so even going
with say -10dB leads to the conclusion AB will better than B.
3) The examples shown in Self's books IMO use the worst possible
output stage configuration, the CFP. Elsewhere in his books Self
shows the CFP has the lowest large signal distortion, but also
shows buried somewhere that the EF has lower distortion than
the CFP for the first watt.
This is because the CFP is all over the place in the 1st watt,
whilst the EF for the first watt tranverses a smooth peak.
When the two characteristics are tranferred to class AB, the
CFP gives a far more defined step in the output stage gain
than that which would occur with an EF output stage.
The sharper the step = the more higher harmonics you get.
4) IMO for a 100W amplifier really clean distortion performance
up to 10w is far more important than low distortion at full power.
Even at full power (unclipped music) most of the time the output
will be well below 10W, surely it is this that matters most ?
5) For his class A amplifier Self shows that the lower the emitter
resistors the better regarding the AB A to B transition. Useful.
So basically the above points are why I chose the DXamp to build
a small 1/4 A class AB amplifier. EF output, no emitter resistors.
I'll have to reduce the rail voltages somewhat and fit somewhat
bigger heatsinks, but IMO no way will it be inferior to class B.
rgds, sreten.
Thought I'd share some thoughts on class AB.
(I'm making a small 1/4 A AB version of the DXamp.)
No-one has been more responsible than Douglas Self for giving
Class AB a good kicking, so I make no apologies for using
only his examples in my arguments, for reference.
I'll also stick to 1/4 class A, half voltage/current he used.
1) Class AB is worse than Class A and Class B ?
Yes it will sound worse than class A. But worse than class B ?
Just because distortion at near full power is worse than class B ?
Surely that is a non-argument, the pro's and con's not discussed ?
2) 1/4 A AB below -6dB is to all intents and purposes Class A.
Self quotes a typical volume setting of -15dB, so even going
with say -10dB leads to the conclusion AB will better than B.
3) The examples shown in Self's books IMO use the worst possible
output stage configuration, the CFP. Elsewhere in his books Self
shows the CFP has the lowest large signal distortion, but also
shows buried somewhere that the EF has lower distortion than
the CFP for the first watt.
This is because the CFP is all over the place in the 1st watt,
whilst the EF for the first watt tranverses a smooth peak.
When the two characteristics are tranferred to class AB, the
CFP gives a far more defined step in the output stage gain
than that which would occur with an EF output stage.
The sharper the step = the more higher harmonics you get.
4) IMO for a 100W amplifier really clean distortion performance
up to 10w is far more important than low distortion at full power.
Even at full power (unclipped music) most of the time the output
will be well below 10W, surely it is this that matters most ?
5) For his class A amplifier Self shows that the lower the emitter
resistors the better regarding the AB A to B transition. Useful.
So basically the above points are why I chose the DXamp to build
a small 1/4 A class AB amplifier. EF output, no emitter resistors.
I'll have to reduce the rail voltages somewhat and fit somewhat
bigger heatsinks, but IMO no way will it be inferior to class B.
rgds, sreten.
Last edited:
Hi
Most amplifiers are AB because you simply cannot get rid of crossover distortion in a Class B amp. Self's Blameless is a Class AB. It just means Class B with a small quiescent current. It can also mean, as you seem to conclude, a highly biased Class AB where the quiescent current is a substantial value.
You won't get a stable quiescent current without emitter resistors- only thermal runaway instead, and phut! on your output trannys.
The issue is whether you can hear the effects of crossover distortion. In a Class AB (perhaps we need a transistor version of AB to designate whether the quiescent current is high or low. In valve terminology AB1 was no positive grid bias, AB2 was positive driven grid. Maybe ABH for high standing current and ABL for low?) there is a transition between A and B at high currents. Calculate the open loop gain for each case and check if you could discern the difference in closed loop. Typically, a classic amplifier design using input (single or differential) - VAS- complementary drivers and output pair often tied the base resistors of the outputs to mid-rail. Therefore, when the output transistors turned off the gain of that stage drops by the loss of the output transistor gain. This could mean the difference of an open loop gain of say 2000 to 4000, but of course you hope the other side is then operating. In an ABH design you will have this open loop gain shift, and distortion as a result.
The ABL design with complementary EF stage avoids the issue by connecting the bases together so that the drivers (ideally) remain in Class A. Only the outputs operate in B, but things are never quite so ideal in reality.
John
Most amplifiers are AB because you simply cannot get rid of crossover distortion in a Class B amp. Self's Blameless is a Class AB. It just means Class B with a small quiescent current. It can also mean, as you seem to conclude, a highly biased Class AB where the quiescent current is a substantial value.
You won't get a stable quiescent current without emitter resistors- only thermal runaway instead, and phut! on your output trannys.
The issue is whether you can hear the effects of crossover distortion. In a Class AB (perhaps we need a transistor version of AB to designate whether the quiescent current is high or low. In valve terminology AB1 was no positive grid bias, AB2 was positive driven grid. Maybe ABH for high standing current and ABL for low?) there is a transition between A and B at high currents. Calculate the open loop gain for each case and check if you could discern the difference in closed loop. Typically, a classic amplifier design using input (single or differential) - VAS- complementary drivers and output pair often tied the base resistors of the outputs to mid-rail. Therefore, when the output transistors turned off the gain of that stage drops by the loss of the output transistor gain. This could mean the difference of an open loop gain of say 2000 to 4000, but of course you hope the other side is then operating. In an ABH design you will have this open loop gain shift, and distortion as a result.
The ABL design with complementary EF stage avoids the issue by connecting the bases together so that the drivers (ideally) remain in Class A. Only the outputs operate in B, but things are never quite so ideal in reality.
John
Correction - the output bases are connected through a resistor, not actually tied together, that would be Class B!!!
I agree with the opening post almost entirely word by word.
Class A is preferred for low power because of lower distortion which measures as a percentage proportional to the actual output power
Class B is preferred at high power simply to save on huge components and the cost of electricity. ( As far as it goes I personally believe that the HUGE single ended class A amps that give a minimal number of watts output are very severely misguided).
At higher powers a person listening in a room will be hearing greater distortion from a whole host of reasons, such as physical ears, reverb echo and resonance, and simple basic increased speaker distortion with increased cone excursion and at the lower high power frequencies, magnetic and thermal speaker coil effects. ie. At high power ultra low amplifier distortion becomes far less critical.
Class AB makes a lot of sense. Not AB as in 5 to 50mA of standing current simply to optimise the crossover distortion, but a proper significant standing current of say 250mA to 1A so as to ensure class A at normal listening volume.
Very shortly after JLLH published his original class A amplifier in 1969 ( a classic example in it's first iteration of poor electronic design in that standing current depended on the appalling low beta of prehistoric power transistors). He published a much less known design for a very novel class AB amplifier with similar excellent performance. I have taken this design and over the past 25 years modified it quite a few times to produce versions that use modern transistors, tailored to different powers and transformer secondary voltages and speaker impedances. In the process I upped the standing current into a range of 500mA to 1A. and both in computer simulation and distortion analyser test it performs lovely. In the process I also stabilised the standing current better by replacing the bias resistor with a conventional amplified Vbe bias circuit.
I only went to all this trouble because I found listening to the various versions was also lovely.
There are some up and down sides. It is a single supply design and has a large electro cap in the output line, but this and the necessarily larger heatsinks obviates any need for overload circuitry. Single supply amps are relatively poor in common mode supply rejection and since the supply is delivering significant current at even very quiet volumes then the power supply smoothing is a serious subject in its own right ( Class B amps do not have this problem). On the upside the overall design is extremely simple using only 6 transistors especially if modern 3281/1302's are used.
BTW In passing I have used cfp's frequently in signal amplifier stages, however in the output stage of a power amp they are nothing but trouble and the measures often needed to stabilise them have a negative effect on overall performance.
Class A is preferred for low power because of lower distortion which measures as a percentage proportional to the actual output power
Class B is preferred at high power simply to save on huge components and the cost of electricity. ( As far as it goes I personally believe that the HUGE single ended class A amps that give a minimal number of watts output are very severely misguided).
At higher powers a person listening in a room will be hearing greater distortion from a whole host of reasons, such as physical ears, reverb echo and resonance, and simple basic increased speaker distortion with increased cone excursion and at the lower high power frequencies, magnetic and thermal speaker coil effects. ie. At high power ultra low amplifier distortion becomes far less critical.
Class AB makes a lot of sense. Not AB as in 5 to 50mA of standing current simply to optimise the crossover distortion, but a proper significant standing current of say 250mA to 1A so as to ensure class A at normal listening volume.
Very shortly after JLLH published his original class A amplifier in 1969 ( a classic example in it's first iteration of poor electronic design in that standing current depended on the appalling low beta of prehistoric power transistors). He published a much less known design for a very novel class AB amplifier with similar excellent performance. I have taken this design and over the past 25 years modified it quite a few times to produce versions that use modern transistors, tailored to different powers and transformer secondary voltages and speaker impedances. In the process I upped the standing current into a range of 500mA to 1A. and both in computer simulation and distortion analyser test it performs lovely. In the process I also stabilised the standing current better by replacing the bias resistor with a conventional amplified Vbe bias circuit.
I only went to all this trouble because I found listening to the various versions was also lovely.
There are some up and down sides. It is a single supply design and has a large electro cap in the output line, but this and the necessarily larger heatsinks obviates any need for overload circuitry. Single supply amps are relatively poor in common mode supply rejection and since the supply is delivering significant current at even very quiet volumes then the power supply smoothing is a serious subject in its own right ( Class B amps do not have this problem). On the upside the overall design is extremely simple using only 6 transistors especially if modern 3281/1302's are used.
BTW In passing I have used cfp's frequently in signal amplifier stages, however in the output stage of a power amp they are nothing but trouble and the measures often needed to stabilise them have a negative effect on overall performance.
Last edited:
Hi,
an optimally biased ClassAB with 0r0 as the external emitter resistor should perform very well. Indications of the improvements in distortion performance show as Re is reduced from 1r0 to 0r1.
Just a pity that AKSA suggests that best sound comes from a fairly high Re value of 0r47.
The downside of 0r0 is that the optimal bias can only be identified by measuring the output, either with a distortion meter or with the ears. Unless one knows how to measure the internal 26mV dropped across the effective resistance of the emitter.
Could this be extended to multiple pairs and still hold Re to 0r0? How well matched would the output devices need to be?
I have thought of using a 10pair MJE15034/35 output stage, using Re=0r15 for device bias = 150mA and total output bias of 1.5A, giving ClassA output peak current of 3Apk for 36W of ClassA into 8r0. This would be ~-5dB ref max ClassAB power into 8r0 if using a 70Vac centre tapped transformer. A device bias of ~85mA would give that 10W of ClassA target.
an optimally biased ClassAB with 0r0 as the external emitter resistor should perform very well. Indications of the improvements in distortion performance show as Re is reduced from 1r0 to 0r1.
Just a pity that AKSA suggests that best sound comes from a fairly high Re value of 0r47.
The downside of 0r0 is that the optimal bias can only be identified by measuring the output, either with a distortion meter or with the ears. Unless one knows how to measure the internal 26mV dropped across the effective resistance of the emitter.
Could this be extended to multiple pairs and still hold Re to 0r0? How well matched would the output devices need to be?
I have thought of using a 10pair MJE15034/35 output stage, using Re=0r15 for device bias = 150mA and total output bias of 1.5A, giving ClassA output peak current of 3Apk for 36W of ClassA into 8r0. This would be ~-5dB ref max ClassAB power into 8r0 if using a 70Vac centre tapped transformer. A device bias of ~85mA would give that 10W of ClassA target.
Last edited:
Class AB?
Not many class AB ( That's optimum bias or Self's class B actually) designs handle crossover well so all get branded as bad. Not so IMHO. Some still sound B... teriffic, but also require a lot of skill with the design and execution.
Maybe a bit more R&D will be worthwhile.
'Sounds like a better plan to me. If you are actually seeking class A performance from a low budget class AB design, I think it's just avoiding the underlying issue that you don't like crossover transitions, so use a toaster to avoid or shove them up the power scale to a point you won't notice most of the time. Call it high bias or whatever semantic handle but the method is clearly class A.
Not many class AB ( That's optimum bias or Self's class B actually) designs handle crossover well so all get branded as bad. Not so IMHO. Some still sound B... teriffic, but also require a lot of skill with the design and execution.
Maybe a bit more R&D will be worthwhile.
Last edited:
I seem to remember that as part of Doug Self's excellent work he identified with very good and thorough evidence that the optimum value for RE always came to a value of approx. 0.22R,( in an amp of conventional topology) ie. the value that we all traditionally used in any case.
Using an RE of nowt or nearly nowt is a poor strategy especially in a multiple output device amp. The devices would require further derating beyond that otherwise necessary to avoid secondary breakown/SOAR?safe operating area. For those unfamiliar, there is nothing clever about SOAR, as silicon heats up its intrinsic resistance drops quite dramatically and any hot spot tends to hog current until the whole job fries in little hot spots. Multiply this by competing parallel devices and the problem grows.
An emitter resistor provides negative feedback that assists in stabilising currents and avoiding the problem.
Thanks for the transistor tip - I was looking for a device I could bias to 150 mA for a headphone amp, and this one seems to fit the bill. Much appreciated!
Hi Andrew
Can you point me to a source for these indications?
I always thought that this degeneration reduced the effect of changing gm. Is it possible that although reduced (with some degeneration) the resulting distortion is more objectionable?
Thanks
-Antonio
I dont understand the need for massive amounts of bias current.
I get away with as little as 10mA and the amps sounds great with zero crossover distortion even at high drive levels.
I get away with as little as 10mA and the amps sounds great with zero crossover distortion even at high drive levels.
Hi, no its not in Self's terms, in his terms class B is optimally biased, rgds, sreten.
I'd prefer class aB for optimal bias and class AB for any higher bias.
Self uses B and AB for the above two cases, and poorly IMO class C for B no bias.
Class C as used in radio circuits is very different to class B no bias.
Last edited:
Hi, your saying the DXamp in standard form cannot work ? rgds, sreten.
It does work, the only "Re" is that that is intrinsic in the output devices.
Last edited:
Hi,
Whilst optimum bias is not the subject of my post I agree, as you reduce Re
optimal current increases because as Self says, optimum bias is the voltage.
So you can arrange for the EF pair, but not the CFP pair, in multiples a very
significant class A standing current, and there is no reason not to do it.
An interesting question is what class A standing current is supported by
the required heatsinking for class B. It would seem churlish for an EF output
not to reduce the Re value up to this point, instead of multiple paralleling.
rgds, sreten.
Hi,
No. Self has settled on 0.1R as the optimum value, the implication being
say for triple output devices the effective Re is 0.033R with no issues.
He declined to comment on using 0.033R with single transistors.
rgds, sreten.
Hi,
Well that depends on output stage topology, 10mA might be OK
for a CFP output depending on Re values, but not an EF output.
In a well designed amplifier the distortion is c/o distortion mainly,
there is no such thing as zero c/o distortion, its always there.
rgds, sreten.
In a poorly designed amplifier, other stage distortion may
dominate the figures, but not the spectra, c/o is the worst.
Last edited:
Agreed about c/o distortions. The idea is to have a constant Gm vs Id function. Through the crossover region where both devices are on, total Gm is the sum of the two. If bias is too low, Gm will droop through the c/o region and c/o distortions will occur. If bias is too high the sum Gm of the two devices will be too large, known as Gm doubling, and crossover distortions will occur. BJTs have very large fairly linear Gm vs Ic all the way down to near cut-off and thus requires smaller bias, 20-40mA. Mosfet Gm droops significantly near cut-off and require higher bias, ~100-200mA, in order to maintain even Gm vs Id between the two devices, otherwise the overall Gm droops and c/o distortions abound. These are some important things that determine the proper bias for any class AB output stage. Ballast resistors tend to reduce c/o distortions in addition to providing a little local feedback. Global feedback succeeds in pushing these distortions to higher orders of frequency but they are still there. Distortions above the audio band do play a role in the 'sound' of an amp.
Hi,
I agree with the basics of what you are saying but not the numbers.
Bias is voltage, not current, Re determines the required current.
(For CFP and EF stages the currents are very different.)
I'm talking BJT's, FET class AB is sensibly the only choice.
(For FET's aB by me or B by Self does not really exist.)
rgds, sreten.
Last edited:
I do think we have wandered off due to misunderstanding and confusion between proper class AB and the marginal AB biasing required for satisfactory operation of class B. If I may restate...
In a class AB amplifier ( a proper one ) the standing bias current is at a relatively high level. This is so as to operate entirely in class A at normal listening levels but provide a margin of operation in class B to deal with very high volume levels, large transients above the nominal level and dips in speaker impedance through the frequency range. No attempt is made to optimise for "crossover distortion", instead it is accepted that at a certain high power level when operation moves from class A to class B then there will be a step in the transfer characteristic of the output stage with an attendant small increase in overall distortion. At full power it is accepted that a class AB will have a poorer measured performance than either a class A or a class B amplifier, but that at normal listening levels of power will provide the benefits of class A, ie. zero crossover distortion, but without incurring the power dissipation and component penalties of such power hungry class A beasts.
As such it is a technique only appropriate to domestic amplifiers of say 10 to 30 watts power.
Have class AB amplifiers been unjustifiably maligned in the past and consequently ignored?.........YES
Would you like to share the source of this definition; Is it not in error using Douglas Self's redefinitions, as it seems? I think in any case you'll find the transition is from class A to AB whilst B (unbiased) is never used in traditional audio amplifiers or terminology since germanium days. Self's redefined class B is a condition where only optimal bias is used. That's proper AB in forum terms, as I read them.
Otherwise, we can't tell what you're agreeing about. 😕
- Status
- Not open for further replies.