Can you help understand limitations of ClassA/G ?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I'm wondering about the benefits of operating ClassA with low power rails and then using power-rail tracking to handle high signal levels. Most of the time I would anticipate operating within ClassA.

I put together a simple simulation to start with.

This kind of circuit is a bit of a mystery to me, and reading the threads that came up in a Search didn't shed much light on things for me. Does anyone have experience with kind of thing ?

Here's the schematic (one version is a reference ClassAB design, the other the ClassA with rail-tracking)...
 

Attachments

  • schem.gif
    schem.gif
    28.1 KB · Views: 759
Here's what the output looks like

-with two different signal levels.

One signal is small and stays within the range of the ClassA stage, the other signal is larger and pushes the supply rails to track the signal.

Signal at output = green trace
Inner positive rail = red trace
Inner negative rail = blue trace
 

Attachments

  • out.gif
    out.gif
    18.5 KB · Views: 685
The FFT 'distortogram' of the outputs at the two signal levels

ClassAB reference = blue traces
ClassAG = green traces

Here's the rub - even when rail tracking isn't needed, the output from the ClassAG amp still shows artifacts from the rail tracking circuity. If I 'disconnect' the rail-tracking feed so the ClassA amp section is operating independently then the FFT is clean.

So I conclude that the rail tracking injects too much 'rubbish' into the inner power rails. Somebody else has surely seen this before - is rail-tracking flawed or is there something I'm missing :confused:
 

Attachments

  • fft.gif
    fft.gif
    15.6 KB · Views: 682
Bigun said:

So I conclude that the rail tracking injects too much 'rubbish' into the inner power rails. Somebody else has surely seen this before - is rail-tracking flawed or is there something I'm missing :confused:


Originally posted by Lumba Ogir
To put it mildly...

I think it's possible to do class G in such a way it won't have any significant negative impacts when it's not being used. So I'm not sure what's going on in your simulation.

The hard part with class G is what it does when you ARE using the added rails. All sorts of issues come up then like what's known as "rail commutation" which is limited by the speed of the power diodes. Douglas Self has written fairly extensively about class G and has both simulated and measured actual designs that have similar distortion to conventional class B designs when below the class G threshold--i.e. below 0.001% at 1khz.

The above is proof the rails need not "inject" or hurt anything when they're not being used. The performance of his class G design, below the switching threshold, is very similar to his optimized class B "blameless" amplifier. While I don't agree with everything Self writes, it's hard to argue with his published simulation and actual measurements of class G amps. And class A shouldn't change anything.
 
The advantage, to me at least, is a very high quality class A amp for 90+% of one's listening that doesn't heat half the house but still has enough power for listening loud when you want to. High power class A designs are horribly inefficient. They're nice in the winter in cold climates but only if you don't have a cheaper source of heat.

Crossover distortion, in a well designed amp, is typically the biggest distortion left. And the only real way to get rid of it,is Class A. So it depends on if you care about small amounts of distortion.
 
AKSA said:
Reduce dissipation in the output devices, only to throw it all away in the commutating devices?Hugh

When the amp operates within the range of the inner rails it is pure ClassA with some additional wasted power in the diodes used to protect the inner rail power supplies but these have a relatively small voltage drop across them so not much to worry in terms of commutating losses. The outer rails provide no current except for the idle current flowing through the ClassG driver devices.

When the inner rails can't cope, then the current has to flow through the additional BJTs that feed current to the inner rails and dissipation will climb much higher - but this only happens on transients or when there volume is cranked. The idea here is not to make a PA amp but to get the higher quality available from operating in ClassA for most of the time and yet have some headroom.

As I see this, it should in principle be possible to achieve real ClassA output over normal listening levels (for me) without excessive power dissipation because the inner rails are low voltage. And when it moves into ClassG at higher levels I'm not much worse off than ClassB.


RocketScientist - you echo my thinking; the million$ question is how to clean it up in order to realize this potential ?




:D
 
Bigun said:

RocketScientist - you echo my thinking; the million$ question is how to clean it up in order to realize this potential ?
:D

If you haven't found it already, you should start at http://www.dself.dsl.pipex.com/ampins/classg/g.htm There's lots of info there (and on his site in general). If you want even more, the 5th edition of his book might be worth buying. There's lots of other info out there, but Self is one of the few to cover class G in a fair amount of detail. He also posts here on DIYA once in a while.
 
RocketScientist,
basically, I agree with you, however, there`s no reason to fool ourselves: switching is the cause of the most unpleasant distortions, depending on device type (diodes are worst). Only class A is free from switching, consequently, increasing efficiency in any way means switching.
It`s much harder to agree with Self`s distortion figures of (optimally biased) output stages, emitter resistors do not reduce crossover distortion.
 
Lumba Ogir said:
RocketScientist,
basically, I agree with you, however, there`s no reason to fool ourselves: switching is the cause of the most unpleasant distortions, depending on device type (diodes are worst). Only class A is free from switching, consequently, increasing efficiency in any way means switching.
It`s much harder to agree with Self`s distortion figures of (optimally biased) output stages, emitter resistors do not reduce crossover distortion.


AKSA said:

....... or non-switching Class AB, if you can arrange it.......

There is no such thing as "non-switching Class AB" once you exceed the Class A level it starts switching. You could have just one set of rails and crank up the bias to run say 15 watts in Class A. This would be a classic "Class AB" amp. But because of the high bias current, and high rail voltages, the amp would be running very hot at idle.

To compare the idle power (dissipation) of 4 different stereo amps that can deliver 200 watts/ch into 8 ohms:

200 watt Class B: <30 watts
15 Watt Class A/200 watt Class G (AG): 70 watts
15 watt Class A /200 watt Class B (AB): 250 watts
200 watt pure Class A: 900 watts

So the Class AB amp uses almost 4 times as much power just sitting there doing nothing compared to the Class G version. Both are free from crossover and "switching" distortion up to 15 watts.

I've seen similar distortion numbers to Self's using an Audio Precision with the LME49810 and LME49830 driver chip. I've also seen what happens when you overbias BJT outputs. The gm doubling you get when a stage is overbiased and transitions from Class A to Class B is real.

As you turn the bias up past a certain point, the distortion RISES if the power level is high enough to swing into Class B and the amp has low enough distortion to see the effect. With MOSFET outputs, however, it's a different story. Yes, it's easier to just crank up the bias than build a Class G amp, but see above. It will use 4 times the power at idle.

I agree crossover distortion is bad. That's why I like Class A but the efficiency is horrible. If I understood Bigun, he mostly expects to listen at low power levels. And I don't think he cares much about a little extra distortion when he wants to listen at loud levels. So, for his needs, Class A up to say 15 watts and then Class G is a very good solution.

I think 98% of the DIYers either don't build amps with low enough distortion and/or don't have the equipment to measure many of the sorts of distortions Self has documented. Just getting a ground slightly wrong or a thousand other things can cause more distortion than we're talking about here.

Lots of DIYers mostly go by their ears. They turn the bias up expecting it to sound better, and not surprisingly, they think it sounds better. But if you have a really low distortion amp, and the equipment to measure it, I think many would find more bias isn't always better.

Still others here seem to mostly use simulation to get their data. And, again, for very small levels of distortion it's just not accurate enough. The models of transistors, etc. are only estimates, and they're incomplete. It's rare that a power amp will measure in real life anything close to what the simulation predicts with respect to small levels of distortion.

I have seen the distortion get worse past the optimum class B bias point--exactly as Self documents. But I've only seen it on amps that already have really low distortion and by using equipment that can measure very low values of THD like an Audio Precision. On an amp that has a THD+N "floor" of say 0.01% at 1 Khz, I doubt you'd ever see the gm doubling effect as it would be well below the other distortions.

And emitter resistors do reduce distortion. They provide local feedback to the output transistors and cause them to be more linear. But the only real cure for crossover distortion is Class A and even Self admits that. Lots of methods have been tried, including Self's new "Class XD", but I'm not convinced the benefits outweigh the costs. I think many have been more marketing hype than real progress. Self's data clearly shows even his new Class XD doesn't perform as well as Class A and it also compromises efficiency.

So it's all a compromise. The Class A/G amp Bigun is proposing will have slightly more distortion above 15 watts than a Class A/B amp, and will cost a lot more to build, but it will be far more efficient.

A pure Class B amp can have less distortion above 15 watts than either the A/G or A/B amp but it will have more distortion below 15 watts.

And a pure Class A amp can beat them all at all power levels but wastes massive amounts of expensive electricity and would be very large, heavy and costly to build if you wanted anything close to 200 watts.
 
Originally posted by RocketScientist If you haven't found it already, you should start at http://www.dself.dsl.pipex.com/ampins/classg/g.htm

Thanks - a handy link indeed. Interesting to know that Self sometimes posts into this Forum.

from D Self: Class-G is now out and about [...] you can read all in Electronics World for Dec 2001 and Jan, Feb 2002. There is a fully worked-out design, and even a PCB available to ease building it.


It seems from this link that all the answers to my questions could well have been published in EW a few years back - does anyone here by any chance have that article ?

For me the jury is still out on non-swtiching AB. I've seen a lot of people enjoy the sound of ClassA and ClassAB. I've seen very little in the way of non-switching and very little in the way of ClassG posted here.

For those addicted to Spice, like me, I attach the simulation file. To use the file, change the file extension from .txt to .asc (you may have to go into Windows folder options to unhide extensions first)
 

Attachments

  • harry77_classgh.txt
    16 KB · Views: 54
RS,

You may be wrong. It is a loose term, I agree, but it describes what looks to be a push pull circuit with relatively low bias wherein none of the output or driver devices ever switch off. That is, they do not drop below the quiescent, or a reasonable proportion of the quiescent. This gives the principal benefit of Class A, non-switching outputs, without the high dissipation with all its attendant issues.

Say not how it can't be done, but rather how it CAN.

Examine Steve Dunlap's circuit, for starters.

Hugh
 
RocketScientist said:


There is no such thing as "non-switching Class AB" once you exceed the Class A level it starts switching...


United States Patent 3,995,228
Pass November 30, 1976

--------------------------------------------------------------------------------
Active bias circuit for operating push-pull amplifiers in class A mode


Abstract
A transistor audio amplifier of the Class AB complementary symmetry type is biased with a new bias circuit to keep both transistors on during all normal signal conditions. The bias circuit includes a pair of series connected V.sub.BE Multipliers with an interconnection between the two multipliers receiving signal feedback from the amplifier output such that the Multipliers can bias both complementary transistors on for all normal signal operating conditions.


"re-invented":

New Biasing Circuit for Class B Operation
Switching distortion is generated in class B operation because an input signal above a certain level causes the two transistors performing the push-pull operation to switch off completely at the same time. It is discussed how complete switching off of the transistor can be avoided and how the off state can be transferred smoothly to the subsequent on state. Ultimately it was found how these effects can be achieved simply by adding a pair of transistors to the bias circuit.

Author: Tanaka, Susumu
Affiliation: Sansui Electric Company, Ltd., Tokyo, Japan
JAES Volume 29 Issue 3 pp. 148-152; March 1981


http://www.aes.org/e-lib/browse.cfm?elib=3926
 
In another thread (Harry77) I am exploring non-switching AB. It's not a good description calling it non-switching since holding a transistor just-on is not the same thing as keeping it active (in my view) but it's an interesting option.

But here, I want to know if it's possible to make ClassAG work and work extremely well ?
 
just pointing out that "non switching" and "sliding bias" are terms of the audio amplifier design art with more behind them than a non-specialist would construe from a superficial semantic analysis



On Class G

texas instruments seems to think so:

http://focus.ti.com/lit/ds/symlink/ths6032.pdf

psrr from Vhi is 20 dB worse than Vlow - the opposite of your circuit?
fig 13-15 show distortion vs ps levels

doesn't show output equivalent circuit though


john
 
Hugh,
many hopeful attempts have been made, in the end doing more damage than good; in practice by the means of an additional error amplifier. The problems are many, accuracy (in time), for instance. Both input voltage and output current need to be supervised.

RocketScientist,
again, I agree with you.
And emitter resistors do reduce distortion.
They do, in more than one way, but not crossover distortion (as stated).
 
First, just to be clear, I agree with Lumba Ogir that crossover distortion is the worst distortion left in a well designed Class B amp. And I mostly agree with Lumba's statement: "Only class A is free from switching, consequently, increasing efficiency in any way means switching." Some of the fancy Class A bias schemes do claim to increase Class A efficiency without introducing switching. But more on those below...

What Bigun is proposing is a pure Class A non-switching amplifier up to some desired power level. I used the value of 15 watts above, but it could be any power level. The extra class G circuitry should not degrade the performance of the amp when it's used below 15 watts. It should perform just like a dedicated 15 watt Class A amp with a single set of power supply rails.

But, for many, 15 watts isn't enough. So what do you do for more power? There has to be some compromise in either efficiency or distortion to get more power.

I could have added a 5th kind of amplifer to my list above and called it "Magic Bias" ;) As I said, there have been many attempts to try and get Class A performance without the full efficiency penalty of regular Class A. There are sliding bias schemes, dynamic current sources, bias offsets, and lots more. I think most were invented (or "re-invented" as pointed out above) for marketing reasons. But the schemes I'm aware of all suffer from one or more of the following problems:

- They don't achieve Class A performance (i.e. there is still some form of crossover distortion left or the method adds new forms of distortion). Even one of the newest of these, Self's/Cambridge Audio's new "Class XD", suffers this problem.

- They reduce efficiency so much from Class B that some start to approach a Class A amp. Or they just adds lots of circuitry and only provide a slight real world improvement over a pure Class A amp. In either case it seems better to me to remove all the extra circuitry and just run class A.

- They are very costly to implement or have other reasons why they are not generally not used.

If there was some practical method for solving crossover distortion without much efficiency or other penalty, either all the high-end amp designers would be doing it, or whichever company held the patent would be dominating the market. Neither has proven true. So it would seem none of the "Magic Bias" schemes have really caught on beyond being used as marketing hype by some manufactures.

As Lumba said: "many hopeful attempts have been made, in the end doing more damage than good" I agree. It's like the Mazda Rotary Wankel engine. It solved some problems but created too many new ones to gain widespread use.

So to answer ASKA's request to say how it can be done, I think any approach is a compromise. It really depends on what's important to the owner. That's the cool thing about DIY, you get to make exactly what you want instead of having to choose something off the shelf that isn't quite right.

In this case, I think Bigun's approach has a lot going for it for his particular needs. He could end up with a very powerful amp that for 95% of his listening has the sound quality of a pure class A amp yet idles at only 70 watts (or whatever value). The plus is better sound at lower listening levels and higher efficiency, the minus is higher cost and higher distortion at high power levels.

As for how to do the Class G part, there's an entire chapter in Self's book on Class G. He presents designs, measurements, variations, simulations, etc. And his Audio Precision graphs show there is no penalty to the Class G circuitry when running only on the lower voltage rails.

And, finally, I'm not sure what you mean Bigun by: "For me the jury is still out on non-swtiching AB." To me you can't use the words "non-switching" and "AB" together. They are mutually exclusive. I do agree there are plenty of great sounding Class B amps out there. So if you're saying perhaps crossover distortion doesn't matter as much as some think, that's a whole different debate ;)
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.