The many faces of distortion

[amplifierguru]

hi Kanwar,

consider this.. is where I come from..provided the amplifier has a forward path ( no discontinuity) through crossover , then (nested) NFB will take care of the rest! Sure you can set too low with an audible outcome but, in general, it is unlikely with NFB on your side AND sense to design an acceptably linear output stage optimised for low Iq.

Such an output arrangement is also more tolerant of the vagaries of loading.

Cheers,
Greg

 

[kASD]

hi Kanwar,

consider this.. is where I come from..provided the amplifier has a forward path ( no discontinuity) through crossover , then (nested) NFB will take care of the rest! Sure you can set too low with an audible outcome but, in general, it is unlikely with NFB on your side AND sense to design an acceptably linear output stage optimised for low Iq.

Such an output arrangement is also more tolerant of the vagaries of loading.

Cheers,
Greg

Yes you are rite that NFB takes great care of the rest,
but Greg , what will you do when you parallel 12 mosfets each having gate capacitance 5600pF, then the consideration only goes to the driver stage that is how well the mosfet is being driven ...continuous or discontinuous....mode...

have a nice day ,
Kanwar

 

[amplifierguru]

Oooh kanwar,

You put me through this torture? Did I challenge your religion?no. But still you ask me how to live with 12x5,600p to drive! How to do this and live a frugal ghandian life? I cannot see this as possible.

I would be seeking some alternatives. Besides using individual drivers, have you considered you may have an overkill problem?

Cheers,
Greg

 

[kASD]

Oooh kanwar,

You put me through this torture? Did I challenge your religion?no. But still you ask me how to live with 12x5,600p to drive! How to do this and live a frugal ghandian life? I cannot see this as possible.

I would be seeking some alternatives. Besides using individual drivers, have you considered you may have an overkill problem?

Cheers,
Greg

HAHAHA,

Greg my friend......heheeheee,

Yes i do have overkill problem or you say me as an overkiller..
because there is a reason besides this type of mind set of mine..

In india a local made amp with brandname "Stranger PBT 501" use 16 2N3773 bjt's and is rated at 500 WRMS at 4 ohms [www.strangeraudio.org]but people here in our country use this amp at 2 to 1 ohms application and it drives the load very happily....if any foriegn brand amp or our amp fails this test then it wont stay in the indian market much longer. large majority of people here are not very literate in terms of pro-audio technicals...we manufacture amp rated at 1000W at 4 ohms but people sometimes put our amp with 2 ohms load connection....and our amp has to pass this test.....suppose a sound operator has 4 amps running on a live program with each amp having connected to say 4 ohms then if one of the amp fails due to some reason then the guy would simply put the load of failed amp to another relates to 2 ohms and the show is still running,,, secondly european countries have cold climate and indian summers are at average 55 degree celcius at afternoons , so overkill is safest option...

Using individual drivers solves the problem to much extent....
but i was just joking..we only parallel upto 8 devices having gate capacitance 2400pF each and driver idle current is 100mA

cheers,
Kanwar

 

[amplifierguru]

Ha! I thought as much.

2N3773's! - how can they listen to such an output stage! These, I thought, were relegated to museum pieces when the ubiquitous MJ15003/4 appeared in 1974. I was there.

A good 2 slope SOA and thermal control with not so many output devices and you should be the GURU in your country. Lohan Baba!

cheers,
Greg

 

[jan.didden]

The smaller the emitter resistors, the larger the optimum bias current. For 0.22 Ohm, Self computes 107 mA as the optimum. For 0.1 Ohm, it's about 210 mA. Go enough below optimum bias current and you get the peak at zero that doesn't dip at all. So what you had is essentially an underbias condition for the zero Ohm resistors. 100 mA is close to optimum for 0.22 Ohms, showing the dip around zero. That's why I was wondering about Jorge's bias current too.

It looks like Self's optimum bias condition also correlates with minimum variation of output impedance for a fixed EF configuration with varying bias. Makes sense - just two different ways of looking at the same thing.

Thanks Andy, Jorge, Evil, Guru, etc..

This thread is turning into a mother lode!
I have one question on one of your early graphs Andy. It shows the expected dips, but I noticed that the peaks were very, very sharp. I understand it is also possibly graphics-related, but such a sharp peak would be very difficult to handle by any nfb. If this peak was more gently, the nfb would be more effective. (Very unscientific this statement, but you get the point I'm sure). Now some of the other graphs, I think those from Evil, showed more gentle peaks. I wonder if there is anything in the circuits that let us tweak it for gentle peaks so that the final result has much less higher harmonics?

Jan Didden

 

[kASD]

Ha! I thought as much.

2N3773's! - how can they listen to such an output stage! These, I thought, were relegated to museum pieces when the ubiquitous MJ15003/4 appeared in 1974. I was there.

A good 2 slope SOA and thermal control with not so many output devices and you should be the GURU in your country. Lohan Baba!

cheers,
Greg

2N3773 are available here at 3 pieces for one dollar only....

visit this link http://www.bel-india.com/BelWebNew/UploadedFiles/Products/TO-3 PACKAGE SERIES.pdf

and see the extent of indian semiconductor industry by yourself...old obsolete devices at throw away prices.....

regarding 2 slope SOA and Thermal control.. i say this:

here an amp known as Studiomaster uses Triple slope SOA protection and along with thermal shutdown and no one is happy with it because it usually attenuates the output or shutdown in the event of high temp condition.. and the live program which is running is stopped and operator smashes his head against the wall.....nothing else....

cheers,
Kanwar

 

[Graham Maynard]

Hi Mikeks,

In Post#224 you challenge that my (back-EMF control/distortion related) statements are at variance with all known and proven facts, and that it might be more illuminating if I provide some means of demonstartion.

Did you not see my Posts#174/5 in this thread ?

This example was via a non-overloading steady sinewave input drive, also without any effects from a suddenly starting first cycle investigation !
Had a plain resistor been the amplifier's *test* load it would have shown a considerably lower distortion residual (normally reported as thd in designer specification sheets), and there could/would not have been any reverse commutation in the critical class-AB bias region because there would not have been any back-EMF acting upon the amplifier's NFB loop generated internal inductance !

I am not saying that all amplifiers behave like this, but for you to feel justified in writing that my statements are at variance with and break known and proven facts, shows that something is wrong.

Might I suggest that you try simulating some amplifier circuits with a more realistic load using the old fashioned fundamental nulling technique.

Feedback theory was first published before I was born, and I don't disagree with it !
Hence my statements *are* based upon NFB loop generated/damping factor related errors, so maybe you could re-examine the known/proven facts and show me exactly where you believe I have 'gone wrong' !


It is possible to improve forward measured linearity on a stage by stage basis, and hence reduce measured thd when using a resistor load, but to what end if the resulting amplifier imparts a distinctly individual behavioural characterisation (distortion) upon loudspeaker reproduction as its NFB loop (if non coherent) attempts to control back-EMF ?

Anyone who disbelieves that (immeasurable?) global NFB loop induced effects are clearly audible should listen to a non-global NFB design.

As to which amplifier distortion mechanism is less annoying between 'thd' and 'back-EMF/NFB loop induced', thus often becomes amplifier/loudspeaker combination dependent.


Cheers ........ Graham.

 

[Jorge]

Hi Jorge,

What's the DC collector current of the output stage? My guess is it's pretty low?

Hello, Andy, yes, it was lower than yours - 55 mA.

Here's a plot with 104 mA.

Have you tried to do a sine .tran in Vout and then do a d(v(out)) plot?
This nice feature (it was unknown to me) will show clearly the crossover disto...
It's a pity the models are not real!

For the models I'm using, Iq should be 85mA for no crossover disto.

And fiddling slightly with the emitter resistor of the NPN output (.18 in lieu of .22) the symmetry is improved.

 

[Jorge]

And here's 88mA, Re NPN .18r

 

[bocka]

It's a pity the models are not real!

Why do you think the models are not real? To me your simulation results comes very close to reality. What we see is crossover distortion on a class b amplifier. The only way to decrease it is paralleling multiple devices and increasing the bias when using class b output configuration. I've compared the given models here from the forum and compared it with the datasheets. The models really come close to reality. If you want to improve linearity the only way is using more output transistors as multiple pairs have a lower resistance and in this way the distortion caused by a varying output resistance is reduced.

But this is only one aspect when designing a HQ output stage.

And what's also important real devices differ from part to part. The question is how close any model can represent reality if real devices have large tolerances...

 

[mauropenasa]

Originally posted by Janneman
But I think you lose sight of the objective. You throw out the baby with the bath-water. You can use transimpedance amps to suppress effects of EMF *to the amp*, but that means that the EMF effect is fully remaining at the output and will necessarily intermodulate and interfere with the wanted output signal. What do you want, low EMF suseptibility or good sound?

Hi Jan, your (amusing) reply balance the my (apparently ) little scientific affirmations. I creed to non-being capable of expressing a "decent" concepts in this Language, for which will limit to the maximum my interventions in this forum. In the post that you comment, I have perhaps compact of the errors of definition on account of the translation, but not report me to a transimpedence amp but to a "transimpedance output stage" to use in place of " unit gain output stage "( emitter/source follower ) in a multi-stage NFB circuit. The development of amplifiers with "transimpedance output stage" not only is not a my invention, but is considered the system more "musical" in absolute by the more important world planners. They make use of this technique practically all the traditional valves amps, all the single ended class A, and launch approaches to solid state elaborated by consolidated planners. The recent discussions in this Thread are confirming the theories of anyone uses this techniques, confirming that even from the back_EMF point of view it is able superior being to other topology. You ending asks that turns me would be able have endless replicas, but doesn't interest me polemize excessively...

Ciao

Mauro

 

[Jorge]

a "transimpedance output stage" to use in place of " unit gain output stage "( emitter/source follower ) in a multi-stage NFB circuit.

Ciao, Mauro

A link for such a stage, per favore?

Grazie,

 

[andy_c]

I have one question on one of your early graphs Andy. It shows the expected dips, but I noticed that the peaks were very, very sharp. I understand it is also possibly graphics-related, but such a sharp peak would be very difficult to handle by any nfb. If this peak was more gently, the nfb would be more effective.

Hi Jan,

Only have time for a very quick reply here - working. My early graphs had a +/- 20 A min and max on the current, while Evil's had +/- 8. So his look more spread out. I had to go with +/- 8 on the later MOSFET plots as those weren't even readable at +/- 20. Also, my vertical axis was autoscaled to blow up the full vertical variation, while Evil's showed two plots on the same graph, compressing them vertically somewhat. I could re-plot my orignals on the same scale as Evil's tonight and we could look again. It's hard to mentally adjust the graphics .

 

[mauropenasa]

Hi Jorge,
A variation of this approaches (transimpedance output stage) finds you in the single ended of Pass as Alep family , in JLH 10W class A, in all the valves amps with exit transformer, and in complete way in the Parravicini work for Musical Fidelity like A370, P180 etc... ( my references, from which I have borrowed the input stage... ) and if you want a similar example elaborate by a "incapable" are able see my THread:http://www.diyaudio.com/forums/showthread.php?threadid=54571

Ciao

Mauro

 

[lumanauw]

Hi, Mr.Maynard,

I'm very interested in this :

It is possible to improve forward measured linearity on a stage by stage basis

How to do this? I can think of 2, putting heavy degeneration (RE) on each transistor or changing every transistor to CFP. Is this correct?

 

[Mr Evil]

For comparison, here's the same simulation run on my favourite power amp , two pairs of lateral MOSFETs in the output stage, about 60mA Iq per device, lots of open-loop gain. The graph is the same sort of shape as those done so far, but much lower magnitude.

 

[Mr Evil]

A slightly different measurement this time, on the same Sziklai output stage I used earlier: Output impedance (purple) plotted against time with a 10kHz sine wave current drawn from the output. Similar sort of shape to the DC impedance, but higher in magnitude.

The current (orange) passes through zero a bit later than the impedance dip. Unsurprisingly, this difference in time is roughly the same as the phase shift at 10kHz.

 

[andy_c]

I have one question on one of your early graphs Andy. It shows the expected dips, but I noticed that the peaks were very, very sharp. I understand it is also possibly graphics-related, but such a sharp peak would be very difficult to handle by any nfb. If this peak was more gently, the nfb would be more effective.

Hi Jan,

I'm reading your response again, and I'm not sure now if you're referring to the original BJT circuit or the later MOSFET circuit.

If it was the MOSFET circuit, I'll refer you to the orignal post where I discussed the sharp peaks. The SPICE models have discontinuities in the derivative of the Id vs Vgs curve.

If it's the BJT's you're referring to, I think the plot below will clear that up. I changed my original BJT circuit so it has only a single NPN and a single PNP to make it match Evil's circuit that he plots in post #237. I changed the current sweep from +/- 20A to +/- 8A. I also changed the y scaling of the graph to be exactly the same as Evil's in post #237. I also tried to size the graph so the graticule size matched Evil's as closely as possible. Below is the result. You can see they look very similar now.

You're right about the distortion. The more wiggles in the curve, the higher the order of the polynomial needed to accurately approximate it. And the higher order of the polynomial directly translates to higher-order harmonics when a sine wave is applied.

One interesting thing is that the underbiased condition has less wiggles than the optimum bias or overbias conditions, but the deviation from linear is larger. You'd expect it to have a higher overall distortion, yet have a distribution of harmonics that's weighted more toward the lower-order ones. OTOH, the overbiased condition seems to have the worst of both worlds. It has more wiggles like the optimum bias condition, but its deviation from linear can also be much higher than that of the optimum bias condition. You get both higher overall distortion, and a greater percentage of high-order harmonics too. Yet it's been said that listeners tend to prefer overbias. So much for the "golden ear" types I guess .

 

[jan.didden]

Hi Andy,

Thanks for the clarification. I actually referred to your post 226. (How do you insert a link to a post in your text?). The +/- 20A sclale does make the peaks look sharp, the +/- 8A scale is more gently and perhaps psychologically less offending.

[snip]One interesting thing is that the underbiased condition has less wiggles than the optimum bias or overbias conditions, but the deviation from linear is larger. You'd expect it to have a higher overall distortion, yet have a distribution of harmonics that's weighted more toward the lower-order ones. OTOH, the overbiased condition seems to have the worst of both worlds. It has more wiggles like the optimum bias condition, but its deviation from linear can also be much higher than that of the optimum bias condition. You get both higher overall distortion, and a greater percentage of high-order harmonics too. Yet it's been said that listeners tend to prefer overbias. So much for the "golden ear" types I guess .

Well, since overbiasing goes towards class A, it MUST sound better, no?

Jan Didden