The RC network at the base of Q3 should go at the base of Q2 since in the
actual design it impacts both final transistors and the driver.
Thanks for the correction.
Bryston 3B simulation showing current sharing:
http://www.diyaudio.com/forums/soli...current-sharing-output-stage.html#post5055778
Simulation similar to Kolinummi's figure 7.36, showing that the outputs certainly do turn off:
http://www.diyaudio.com/forums/soli...current-sharing-output-stage.html#post5055859
http://www.diyaudio.com/forums/soli...current-sharing-output-stage.html#post5055778
Simulation similar to Kolinummi's figure 7.36, showing that the outputs certainly do turn off:
http://www.diyaudio.com/forums/soli...current-sharing-output-stage.html#post5055859
it does not matter that they turn off (as in passing zero current) or transition into a constant current state where they no longer control the output.
In both of the above operating conditions one is left with one half of the output stage controlling the output current, when the other half becomes dormant.
Yes it does. A "constant current" flattens the gm doubling region, hence lowers the crossover distortions.
once one half of the output pair stops controlling the output then the other half has to do all the controlling.
In that situation you may or may not get gm doubling.
Whether you get gm doubling does NOT depend on whether the non controlling half is running at zero current or at some constant current. it is not controlling.
In that situation you may or may not get gm doubling.
Whether you get gm doubling does NOT depend on whether the non controlling half is running at zero current or at some constant current. it is not controlling.
This is correct. Static crossover distortion depends on which transistors are contributing transconductance to the output stage, and how much. Ideally, the sum of the transconductances of the top and bottom halves is the same as a function of output current. When this sum of transconductances changes as the net output current to the load passes through the zero region, we have static crossover distortion.
Even if the "going off" side stays on, but only conducts a somewhat constant current, as in some "non-switching" designs, it is not contributing transconductance to the overall output stage, and so does nothing to reduce or influence static crossover distortion. GM doubling is one form of static crossover distortion, where top and bottom are heavily biased at idle, contributing lots of transconductance in the crossover region. When the crossover region is exited, only one side is contributing transconductance, so, in the limit, total transconductance of the stage is halved when the crossover region is exited.
Of course, it is more complicated than this because these changes in gm happen gradually as the output current passes through the crossover region, and the "doubling" of gm in the crossover region is only approaches asymptotically in the limit of very high idle current (which, of course, also leads to a large class A region).
Cheers,
Bob
@AnderwT My point was that the author suggested that the Bryston was, sort of,
non-switching, showed them in his "simulated" diagram as not turning off, when in fact
they do. The author (mainly his diagram) is wrong.
It is true that the Bryston seems to maintain the bias current level even at the zero crossing
of the output waveform as can be seen in my simulation and this has some minor
advantages.
When I think of crossover distortion worth talking about, I'm referring to (probably) an
output stage with class B bias such that the crossover distortion is so bad that it is
audible. I don't think that any of this is audible in an amp with a reasonable amount
of AB bias perhaps even one without global feedback - if it is done right. It is of academic
interest but not very important in my opinion.
non-switching, showed them in his "simulated" diagram as not turning off, when in fact
they do. The author (mainly his diagram) is wrong.
It is true that the Bryston seems to maintain the bias current level even at the zero crossing
of the output waveform as can be seen in my simulation and this has some minor
advantages.
When I think of crossover distortion worth talking about, I'm referring to (probably) an
output stage with class B bias such that the crossover distortion is so bad that it is
audible. I don't think that any of this is audible in an amp with a reasonable amount
of AB bias perhaps even one without global feedback - if it is done right. It is of academic
interest but not very important in my opinion.
The width of that transition region between the OPS halves does dramatically affect the crossover distortion magnitude (and perhaps, more importantly, the high order components). If the biasing regime drags the width of the "turn off" (even if going into a constant current state where it's no longer contributing to the OPS transconductance) out, then the overall distortion is reduced.
This is just a reiteration of what Waly writes in slightly different verbiage.
This is just a reiteration of what Waly writes in slightly different verbiage.
xover in EF2 & CFP
as a relief from dis theoretical stuff, may I ask for comments on 'real life' xover?
In Self's APAD Handbook 4th ed, he compares measured THD with level for EF2 and CFP O/P stages ... Figures 5.39 & 5.40
The EF2 has MUCH less THD at 1W and below though the CFP is superior at full power. He says "This is an unexpected observation, and possibly a new one".
I asked him to investigate this for his new book but got nothing in reply.
Does anyone have his latest edition and can tell us if he did look at this further?
Better still, Bob could you check this out for your new book.
as a relief from dis theoretical stuff, may I ask for comments on 'real life' xover?
In Self's APAD Handbook 4th ed, he compares measured THD with level for EF2 and CFP O/P stages ... Figures 5.39 & 5.40
The EF2 has MUCH less THD at 1W and below though the CFP is superior at full power. He says "This is an unexpected observation, and possibly a new one".
I asked him to investigate this for his new book but got nothing in reply.
Does anyone have his latest edition and can tell us if he did look at this further?
Better still, Bob could you check this out for your new book.
I will be saying a bit more about the CFP output stage in my second edition, and will look for this in some wing spreads and THD sims. However, Doug tends to use very low-value emitter resistors in his CFP output stages, on the order of 0.1 ohm, and biasing to meet the Oliver criteria (which does not really work for a CFP anyway) is tricky, and largely requires starving the output stage in terms of bias.
If you look at his CFP wingspreads, you will see that the crossover region is small, but sharp, suggesting high-order distortion at low signal levels. Because of the tendency for each half of the CFP output stage to have a low output impedance (due to the local negative feedback), the CFP "wants" to do gm doubling. At higher power levels, the effects of gm doubling are less dominant, and the distortion-reducing nature of the local feedback of the CFP begins to come into play by possibly reducing distortion of the CFP output stage under these conditions as compared to a 2EF output stage.
This, at least, is my speculation on why he may have made that experimental observation. With the CFP, even small changes in the idle bias can give substantially different results. I seem to recall that in many of his CFP output stages designs, he is biasing them at only about 10mA. This results in an exceptionally small class A range.
If you don't like gm doubling, but you want enough bias current to provide a decent class A range, don't use a conventional CFP (like Self's). Remember, the First Watt matters.
Cheers,
Bob
I've always been interested in the Bryston...
I don't remember if I did a wingspread on it. If I didn't, I should have...
As I said, I'll have to go back and look at what I wrote.
Hi Bob
Have you looked?
Buffers like 2EF are simple but since the Bryston has gain how did you compare apples to apples, presumably set the gain to unity?
Best wishes
David
The gain is about 3 so I did d(V(vout/3)) and that seems to work, I'll post it when I have a minute.
GASP!!
Is that your polite and long-winded way of saying Self's CFPs are cr*p??!!
I don't like CFP o/p stages for other reasons too but it would be good to find out if this behaviour is just the result of poor design.
GASP!!
Is that your polite and long-winded way of saying Self's CFPs are cr*p??!!
I don't like CFP o/p stages for other reasons too but it would be good to find out if this behaviour is just the result of poor design.
If you have to starve an output stage of bias to keep it from gm doubling, then it is not an output stage that I would advocate using. I am trying to be both polite and succinct
.Cheers,
Bob
How does Class A fit in this description, as a limit of Class AB bias? If I understand correctly, you are saying that increasing the Class AB bias increases the gm doubling effect and hence distortions.
I say the first part is true, but then it also eventually flattens the ridge in what you call "wingspread", which, in turn, after an initial increase beyond the Oliver point, lowers the static crossover distortion. When full class A is reached, the ridge (theoretically) disappears and so are the crossover distortions.
The crossover distortion vs. bias, starting from Class B to Class A, looks like: decrease from a maximum in Class B to the Class AB Oliver point, then increase again (gm doubling with increasing Class AB bias), then decrease again to a flat (theoretically zero) in class A.
It is true that "non switching" is not necessary better for the crossover distortion, since the issue is not turning off half of the output stage, but gm doubling around the crossover.
A relative optimum bias for avoiding crossover distortions is to keep constant a geometrical average of the two output stage halves bias. This has to be valid for all output levels and all frequency of interest. Very difficult to implement, and really not worth the complexity and the potential trouble. I recall Mr. Edmond Stuart tried something in his Autobias designs, using a nonlinear feedback loop, but which proved to be difficult to tame (had a massive cross conduction effect (likely catastrophic in practice) when a fast step response towards the rails was applied at the input). It's also how the LT1166 (if memory serves) bias IC works, but unfortunately it doesn't have the bandwidth to support the whole audio frequency band. Would be ideal for a subwoofer amplifier application, though, but for that we have Class D, no need to bother with anything else.
Hi Richard
I share some of your concerns on the CFP, as we have already discussed.
In particular the -ve feedback is applied around 2 "half" loops so it doesn't necessarily improve the "fit" of the two halves as they try to match at cross-over.
The Bryston is a related OPS that seems to improve on this, maybe a better candidate to put some analysis effort because the CFP looks a bit of a dead end for this application.
The Bryston does seem to work best with two half loops of local feedback, I'm not entirely sure why.
There is thread on this with a link posted earlier but to save you time it's >here<.
Best wishes
David
Chris: then you may well be fine! CFP does run foul during clipping, but doesn't seem like you're biasing for minimum distortion at max power, as Doug does. It does lack the provision to suck out carriers to reduce HF switching loss, unless you run the drivers off a higher rail (which would help with clipping too).