Self is not appealing to an audience, but showing (in simulation) that gm can indeed double. Leach says he is going to show that gm does not double, but then actually shows something different: that loaded voltage gain does not exceed unity - perhaps he thinks these two different statements say the same thing? His graphs thus hide the very effect he is claiming to debunk. I have not slogged through his algebra to find any mistakes, but common sense says that two followers operating in parallel will have twice the transconductance of one. So then the question becomes how they vary with current.OnAudio said:
For a BJT transconductance is proportional to current. For a pair the total transconductance thus is proportional to total current (i.e. the sum of the two BJT currents). Constant transconductance therefore means constant total current - which is means Class A. Class B causes the total to vary, so although he my have shown that gm doubling cannot occur for plain BJT PP output he has also shown that Class B cannot work for this architecture either. Adding emitter resistors provides a solution; he accepts that this makes the equations intractable - and then shows an unhelpful graph.
For a complementary pair gm-doubling can easily occur. My conclusion is that Leach's paper may confirm that gm doubling cannot occur, but only for a circuit which few use (BJT with no emitter resistors) and which will suffer from crossover distortion unless in Class A.
Distortions vary. Some sound horrible. Some are preferred by some people. Hence the presence of distortion does not necessarily mean that all will dislike that amp.
I am still unclear about the purpose of this thread. Others' guesses may be correct.
Well seems everything adds some distortion, some more objectionable than others, we love the distortion introduced by bass reflex designs, good alcohol etc, how objectionable is crossover distortion ?
How objectionable? That may depend on how much there is of it and what kind of source material signal it is distorting. No reason to assume the ear and or brain objectionability response is linear, is there?
What do listening tests say? I seem to recall recently reading that crossover distortion can be a lot less noticeable than people think. I find that surprising but if that is what the tests say then so be it.
PMA picked some violin music for the hi-res crossover distortion listening test. I thought the files were slightly distinguishable blind, but only barely so. Apparently, the one I thought to be slightly less compressed and slightly more detailed was the one with the test amplifier in the circuit. Don't know if the difference to me had to do with crossover distortion or not. I though it might have had to do with buffering between the D/A and A/D in the recording chain.
Also, I'm not sure violin would be the most revealing music for trying to find out if crossover distortion is objectionable. Violin is often first-order modeled as a sawtooth waveform, and it has something of a buzzy sound to begin with.
Also, I'm not sure violin would be the most revealing music for trying to find out if crossover distortion is objectionable. Violin is often first-order modeled as a sawtooth waveform, and it has something of a buzzy sound to begin with.
Something which needs to be remembered is that 'crossover distortion' is a description of a mechanism, not a description of the resultant distortion. Different architectures will produce quite different patterns. To a certain extent, so will different signal levels.
Furthermore, crossover distortion is ambiguous.
There is the gross Xoverdistortion in push pull stages where at Xover both upper and lower leg are OFF.
There is the milder Xoverdistortion where at Xover both legs are ON. Here we trade the both OFF gap for two crossovers: One leg only ON (ClassB), Both legs ON (ClassA), The other leg ON(ClassB). Here some mention gmdoubling, a not so exact term either.
NFB cannot cure this gross Xover distortion.
NFB is the favorite way to reduce this mild Xover distortion.
There is the gross Xoverdistortion in push pull stages where at Xover both upper and lower leg are OFF.
There is the milder Xoverdistortion where at Xover both legs are ON. Here we trade the both OFF gap for two crossovers: One leg only ON (ClassB), Both legs ON (ClassA), The other leg ON(ClassB). Here some mention gmdoubling, a not so exact term either.
NFB cannot cure this gross Xover distortion.
NFB is the favorite way to reduce this mild Xover distortion.
I don't think what you call gross xover happens in class AB; there is no instance that both polarities are off.
Jan
Jan
I never said, gross Xover happens in class AB. As a matter of fact, I said the contrary.
I though I made it clear.
What I call gross Xover is the case of underbias as with BJTs output stage and no base spreader.
BTW, this belongs to class C.
What I call milder Xover is the case of the usual Class AB with adequate base spreader and biasing.
Sorry if there is gross misunderstandings about the term "Crossover". I just try to make it clear.
Last edited:
Class A and D have no audio crossover distortion.
Class AB with Mosfets can equal class A on much less bias ( 75 ma)
Mosfets have soft turn on characteristics and are much faster then BJT's especially turn -off times
Class AB with Mosfets can equal class A on much less bias ( 75 ma)
Mosfets have soft turn on characteristics and are much faster then BJT's especially turn -off times
With class B amps now approaching 1ppm THD+N at full power and any bandwidth, is crossover distortion even significant at these amps? What if the entire distortion product at these low levels were entirely of the crossover variety, would it be audible? In addition to this "truth" exercise, didn't Steve Dunlap solve the crossover distortion problem with his unique output stage?
Last edited:
The current thinking is to measure the zero crossing behavior at very low power, not to judge one way or the other but a class A SET can measure better than a Halcro in some cases.
As for unusual output stages, one should make sure they are robust to the phase angle of current to voltage with real speaker/crossover loads. Circuits that use any amount of "guessing" at the error have a problem with this since the guess usually defaults to the behavior into a purely resistive load with zero phase.
Last edited:
What's a reasonable low level output signal that the analyser can be expected to give sensible answers?
Can't use a normal analyser, either a lock-in amplifier or synchronous FFT techniques. It's actually easily done with a sound card and any of the pre-null techniques like Bob Cordell's distortion magnifier. The signal in needs to be exactly in an FFT bin and average away preferably unwindowed down to the noise floor. At low enough levels you could get away without the null if you have a decent source and use long FFT's to get the equivalent noise BW well below a Hz.
Last edited:
"Class B amps now approaching 1 ppm THD+N at full power and any bandwidth"
Is not possible.
At "Full Power" means that it is clipping and the full power effect is often measured at THD = 1%
Yours
Reodor
I'm sure you're using two different definitions of full power. Why not simply pick a number 100W, 400W both exist.