I can't say I fully see or agree with the points you make in the first few paragraphs. But what you say in your last paragraph is exactly what we are talking about and where the thread started. Class AB where with a high impedance load operation is purely class A and with lower impedance at high power it moves over into the class B region over a proportion of the signal cycle.
Hmm ... a bit off thread?
I'd say that most "class B" amps are AB because of the small quiescent current. So the thread is about a high bias AB mode as I suggested might be called ABH. This will generate a distortion not exactly gm doubling - more like gm halving- when one side cuts out.
I had assumed that the emitter resistors should be chosen to degenerate gm to half so that when both halves of the output stage were on, the overall gm is maintained. I certainly would not use a circuit as in the earlier post without emitter resistors at all, unless you were in fact running this in Class B, or more correctly, a push-pull class C mode. So, to improve stability Re's are recommended. If Ie were 100mA then the nominal gm is 4A/V - and this makes Re=0.25 ohms, which is very close to the figures mentioned.
The difficulty is that while this is true for DC a problem arises in that the dynamic impedances can be different. Also this is only nominal. Actual transistor behaviour depends on whether it is operating in high current injection; whether base resistances and so on affect the effective Re.
IN some other thread I suggested a bias circuit using a true Vbe multiplier. I think that using low - or no- Re's would at least warrant that. In this Vbe multiplier, a CFP are used where the base bias resistors of the NPN first stage divide the voltage in an exact multiple of the number of transistors needed to be compensated. For example, a standard amp. needs four, so the bias resistors might be 3.3k (B-E) and 10 k (B-C). The NPN collector feeds the PNP base. The PNP carries most of the current, and the trim pot adjusts the current in the base shunt path.
John
I'd say that most "class B" amps are AB because of the small quiescent current. So the thread is about a high bias AB mode as I suggested might be called ABH. This will generate a distortion not exactly gm doubling - more like gm halving- when one side cuts out.
I had assumed that the emitter resistors should be chosen to degenerate gm to half so that when both halves of the output stage were on, the overall gm is maintained. I certainly would not use a circuit as in the earlier post without emitter resistors at all, unless you were in fact running this in Class B, or more correctly, a push-pull class C mode. So, to improve stability Re's are recommended. If Ie were 100mA then the nominal gm is 4A/V - and this makes Re=0.25 ohms, which is very close to the figures mentioned.
The difficulty is that while this is true for DC a problem arises in that the dynamic impedances can be different. Also this is only nominal. Actual transistor behaviour depends on whether it is operating in high current injection; whether base resistances and so on affect the effective Re.
IN some other thread I suggested a bias circuit using a true Vbe multiplier. I think that using low - or no- Re's would at least warrant that. In this Vbe multiplier, a CFP are used where the base bias resistors of the NPN first stage divide the voltage in an exact multiple of the number of transistors needed to be compensated. For example, a standard amp. needs four, so the bias resistors might be 3.3k (B-E) and 10 k (B-C). The NPN collector feeds the PNP base. The PNP carries most of the current, and the trim pot adjusts the current in the base shunt path.
John
I am dimly aware of this DX amp, but the circuit as published here is a nightmare of poor loop gain, illusory balancing and a rats nest of various compensation poles, plus an output stage very critical on bias setting that is likely to wander all over the shop in use due to thermal effects. I thought Doug Self had pretty much buried this sort of bad practice.
I don't doubt "some compensation issues"
This is essentially a non existent issue with modern designs.
Gm doubling , switching off of the output devices was a problem
at a time of low beta , low Ft BJTs along with very poor loop gains
attached to poorly designed schematics.
Contrary to some claims, 50mA IQ is no more class B than the same
amp biaised with let s say 0.5A.
It all depends on the devices gains and amp s loop gain to make a 50mA
biaised amp being more class AB than a 0.5A biaised amp badly designed.
As a real world test, i just make a test using a classical low loop gain discrete LIN ,
as the DXamp , and a design using the same Sanken MP1620/MN2488 high Ft darlingtons
as final stage but with with a NE5532as front end in replacement of the differential + CE VAS.
When reducing the bias from 60mA to 0 progressively, the LIN blatlantly
give not only big amount of crossover distorsion, but the sound volume
goes down as well.
With the NE5532 driven OS , nothing of the sort , even when connecting
the two darlingtons bases together , wich increase the dead zone to about
a 2.4V gap.
I agree, it all depends on the design. If you can keep the VAS impedance high then gm doubling/halving does not enter the equation at all. The problem is keeping it high impedance for as high a frequency as possible.
So it all depends on how much influence gm halving has in the particular design considered. I think it is possible to design a good ABH, but depends if the circuit manages the crossover region well. If it can, then I suspect normal AB would work as well. But my preference is class A if you want better.
With a high open loop gain, it is also true that the effects of gm halving are small. But using an op-amp presumably means the op amp is slew rate limited in the crossover region especially if bases are tied together.
John
So it all depends on how much influence gm halving has in the particular design considered. I think it is possible to design a good ABH, but depends if the circuit manages the crossover region well. If it can, then I suspect normal AB would work as well. But my preference is class A if you want better.
With a high open loop gain, it is also true that the effects of gm halving are small. But using an op-amp presumably means the op amp is slew rate limited in the crossover region especially if bases are tied together.
John
Maybe the definition of class A/AB/B in itself, when confined to the amplifier is largely unimportant. What appears to be important however is what the amplifier is driving.
Isn't an optimally biased class AB amplifier the same as an optimally biased class B amplifier providing they are both driving purely resistive loads?
If one then swaps over to a reactive load then whatever definitions you were going by before, both amplifiers now become an optimally biased class AB.
At the same time an optimally biased class AB or class B amplifier would surely be considered a class A amplifier when driving say a 1 meg ohm load, as the biasing required to bring both halves of the output stage cleanly together would exceed the required current to drive said load.
I guess this is all semantics though.
Isn't an optimally biased class AB amplifier the same as an optimally biased class B amplifier providing they are both driving purely resistive loads?
If one then swaps over to a reactive load then whatever definitions you were going by before, both amplifiers now become an optimally biased class AB.
At the same time an optimally biased class AB or class B amplifier would surely be considered a class A amplifier when driving say a 1 meg ohm load, as the biasing required to bring both halves of the output stage cleanly together would exceed the required current to drive said load.
I guess this is all semantics though.
Just starting to understand the issue, AB with high bias and AB with "optimal" bias. But now I'm again confused with why gm effects do not enter the equation.
Can you help me understand this?
Thanks
-Antonio
Ooooh .......
I actually quite like it, you really need to know your amplifier onions
to understand what DestroyerX is up to, it is ideal for high bias AB.
IMO by accident, not design, its good for high bias AB.
rgds, sreten.
Attachments
Last edited:
Hi,
Not to put too fine a point on it, this thread is not about optimum bias if you
read all of it, and quite frankly regarding optimum bias it seems to me to be
attracting those that don't understand the issue in the first place.
rgds, sreten.
sreten
Sorry I can't help (not enough experience with this yet) but I was hoping since you've given this significant thought you could try and explain some of the issues I'm struggling with:
In Douglas Self's Audio Amplifier book he shows a error waveform for the Class AB with higher displacements than in class B (class A is not shown) in figure 6.43 (fifth edition). What I don't understand is why during the presumably one transistor only on portion (the displacement between cross-overs) is significantly higher than the class B equivalent. Beta droop is not enough to explain this based on the class B waveform.
For the same error diagram why don't the plateaus within the positive crossing and negative crossing zero's have the same value?
Lastly what is it about the DX design that makes it more suitable for higher output bias?
Many thanks
-Antonio
Sorry I can't help (not enough experience with this yet) but I was hoping since you've given this significant thought you could try and explain some of the issues I'm struggling with:
In Douglas Self's Audio Amplifier book he shows a error waveform for the Class AB with higher displacements than in class B (class A is not shown) in figure 6.43 (fifth edition). What I don't understand is why during the presumably one transistor only on portion (the displacement between cross-overs) is significantly higher than the class B equivalent. Beta droop is not enough to explain this based on the class B waveform.
For the same error diagram why don't the plateaus within the positive crossing and negative crossing zero's have the same value?
Lastly what is it about the DX design that makes it more suitable for higher output bias?
Many thanks
-Antonio
Absolutely nothing, quite the contrary, it will just fry the power Bjts..
With no emitter resistors , the power devices will quickly
run in thermal runaway and the VBE multiplier , although
mounted in thermal contact with a power device , will do
nothing to prevent such an event , unless it overcompensate
the thermal variations , wich, given the design, is not the case,
as it s rather undercompensating..
Hi magnoman,
I spent a lot of pages in my book (Designing Audio Power Amplifiers) to the issues surrounding output stage biasing and semantics. There is no clean answer to the semantics, but I wrestled with what I thought best and ended up disagreeing with Doug Self. In my opinion, the optimally-biased output stage is class AB, not class B. The terminology I have chosen to go with seems to be more consistent with output stage terminology going back to the vacuum tube days, where virtually all high fidelity push-pull amplifiers were said to be operating in class AB. Class B was understood to have little or no bias at idle and was prone to cause what was then often referred to as notch distortion.
Optimally-biased class AB is generally the idle bias that results in a minimal variation in total output stage gm as the output current transitions through zero current (not necessarily zero voltage - it is the output current that counts). Because the gm of the output transistors is a function of current, the setting of idle bias current controls the gm (1/re) of the output transistor in comparison to the resistance of the external emitter resistor. It turns out that making 1/gm of the output transistor at idle equal to RE theoretically minimizes the distortion due to changes in total gm of the output stage as the output current transitions through zero.
The gm of the output stage is important because under many conditions the output stage acts as it is voltage driven rather than current driven. This is so because over much of the frequency range the shunt feedback of Miller compensation causes the output impedance of the VAS to be significantly lower than the input impedance presented by the output stage. This can be confusing to some because we often think of the VAS as having a very high output impedance. It certainly does at very low frequencies, but it can be surprizingly low at mid-band and higher frequencies in a well-designed VAS with Miller compensation. The voltage drive paradigm of the output stage is especially valid when the output stage is a Triple, where total current gain of the output stage is very high.
Because the output stage is essentially voltage-driven, its transconductance is what counts, and that is why we want gm to be as nearly constant as possible as the output current transitions though zero. If a class AB output stage is heavily over-biased, so-called gm doubling occurs, at least in the limit. This means that with high idle bias, quiescent gm of the output transistors is very high, so re is small. The effective 1/gm of the output stage at idle is then nearly equal to RE/2 (double the gm of 1/RE). At high current, above twice the idle current, only one output transistor is on and the effective output impedance of the output stage is RE (not 1/2 RE, since the other RE is not connected because its associated transistor is turned off). Hence, we have a much larger gm (nearly twice) in the crossover region as far outside the crossover region. This is the inverse of notch distortion caused by under-bias.
Finally, it is better to err on the side of over-bias and gm doubling, since, among other things, this results in a larger class A region where both output transistors are on and contributing transconductance to the output stage.
There are some curves in my book that show how crossover distortion changes as a function of idle bias.
Cheers,
Bob
Hi Sreten,
I have taken your advice and re-read post1.
What is "your ClassB"?
What is "1/4 A AB" amplifier?
I originally understood that post1 was referring to the conventional ClassA, ClassAB and ClassB, but I am now thinking you have changed the definitions.
That unexplained change from the conventional definitions may be causing confusion.
I have taken your advice and re-read post1.
What is "your ClassB"?
What is "1/4 A AB" amplifier?
I originally understood that post1 was referring to the conventional ClassA, ClassAB and ClassB, but I am now thinking you have changed the definitions.
That unexplained change from the conventional definitions may be causing confusion.
magnoman,
Mr. Self probably does not understand it either. He thinks he did find the real culprit behind the output stage distortion anyway: the gm-doubling distortion, easily cured by the "optimal bias", generally valid for all devices in all configurations at all temperatures.
optimal ClassAB bias does not remove nor eliminate crossover distortion.
It merely allows operation of the devices at the minima of distortion in between ClassB and over biased ClassAB.
D.Self makes this very clear. All other competent comentators are confirming the same finding.
There will always be some crossover distortion when operating outside ClassA.
It merely allows operation of the devices at the minima of distortion in between ClassB and over biased ClassAB.
D.Self makes this very clear. All other competent comentators are confirming the same finding.
There will always be some crossover distortion when operating outside ClassA.
If you believe that switching distortion is more tolerable than the gm-doubling distortion then go for class B, no class A overlap.
Hi,
Regarding the last question, EF output topology and no Re's, Vbe thermal tracking.
I'm looking ar figure 6.43 and I cannot understand your question above ....
rgds, sreten.
- Status
- Not open for further replies.