bipolar (BJT) transistor families for audio power output stages

[traderbam]

This thread looks like fun. Can I join in?
Looks like LO needs some support.

Piercarlo,
The problems with global feedback are serious enough even if properly used, noticed by many audiophiles, those who use their ears.

Use of ears - good, good, good!

The time delay is a huge and fundamentally insurmountable problem. There´s every reason to doubt the accuracy of time delay and phase correction techniques.

Yes. You never get something for nothing. Feedback trades accuracy for stability - always. Stability depends upon loop gain, phase shifts and time delays (phase shifts and time delays are different and should be treated separately). And errors can never be eliminated using NFB, only reduced.

Global feedback lowers the measured distortion by transforming (multiplying) harmonics into high frequency dirt, safely outside the measurable range, thus it`s essentially about distortion redistribution.

It is a mathematical fact that feedback multiplies and spreads harmonics in a non-linear system. Not sure this is always "safely outside the measurable range" though.

It also inevitably introduces new distortions.

You'll have to define what you mean by "new distortions". Let's consider non-linear distortions. Feedback in and of itself will not introduce new non-linearities in the circuit unless the feedback components are themselves non-linear. However, being regenerative, the original circuit will have energy in it at different levels and frequencies than it does without feedback and this may excite non-linearities in the circuit that were not previously excited. The extreme case is when the feedback circuit oscillates.

The result is an adverse harmonic spectrum, tones deprived from their harmonic content, a tangle of phase relationships, a clean but unpleasurable sound.

Yes. For example, it is a fact that in some circuits, probably most, too much use of NFB can sap the music of life even though the bench measurements improve. But a more specific cause-effect explanation is required.

 

[SQLGuy]

Use of ears - good, good, good!

Why? And how?

This industry is way too full of people that annouce their conclusions based on totally subjective and uncontrolled listening. I am not saying that the results shouldn't sound good, but I am saying that I have no respect whatsoever for the conclusions someone reaches by listening when they've made no effort to control the testing (by matching levels, by doing blind comparisons, etc).

If the goal is the most euphonic result out of some no-feedback high distortion solution, then, please, listen away, but don't try to make me believe that your result is a more accurate amp whose accuracy can only be verified with your listening tests.

 

[Bob Cordell]

do ClassAB tube/valve amps have the equivalent to Vre?

Hi Andrew,

I don't think so, but if someone worked at it they might be able to come up with an analogous rule of thumb.

In any case, in order for something to be class AB, I don't think it has to be goverened by a Vre rule of thumb. For example, MOSFETs are not governed by such a rule of thumb. They just like more bias.

The Vre rule of thumb basically sets the amount of bias over which you begin to get gm doubling from BJT output stages. To the best of my knowledge, MOSFETs are immune from gm doubling.

One could be tempted to say that exponential devices like BJTs can suffer gm doubling and that square law devices do not, but I'm not sure such a generalization is justifiable.

Cheers,
Bob

 

[Bob Cordell]

Stuff well above 20khz does not matter, it's above the frequency range that the tissues in our ear can pick up, and outside the frequency range that tweeters can play efficiently. Class D takes advantage of that.

Hi Eva,

This is not true. It is another misconception that people believe that it is the harmonics of signals from nonlinearities that bother people. This is just a sinewave view of the world. The big issue is IM components from complex signals that are often folded down into the audio band. This is the sort of thing that causes the spit on a cymbol, for example. The CCIF 19+20 kHz test clearly shows this with spectral lines that lie well below 19 kHz.

THD-20 is merely a way of assessing HF nonlinearity by looking at symptoms. It is like a doctor who diagnoses that you have a disease based on him looking at symptoms and the presence of antibodies, not usually the disease itself (there are, of course, obvious exceptions to this analogy).

Cheers,
Bob

 

[traderbam]

Why? And how?

This industry is way too full of people that annouce their conclusions based on totally subjective and uncontrolled listening. I am not saying that the results shouldn't sound good, but I am saying that I have no respect whatsoever for the conclusions someone reaches by listening when they've made no effort to control the testing (by matching levels, by doing blind comparisons, etc).

If the goal is the most euphonic result out of some no-feedback high distortion solution, then, please, listen away, but don't try to make me believe that your result is a more accurate amp whose accuracy can only be verified with your listening tests.

Always listen carefully.
What do you mean by accuracy? This is a critical question. After all, we don't design amplifiers for the listening pleasure of our oscilloscopes - although some seem to!
What else can one do but listen? Is not the point of hifi to fool the human brain (whose audio perception processing has evolved over millions of years) into imagining it is listening to the original sound source. So unless you have an alternative that is better at perceiving the accuracy of reproduced source as if it were a human ear-brain (and this would necessarily have to be judged by listening tests), then you are best off using a real brain as the final arbiter, preferably your own if you want to enjoy the hifi yourself.

 

[ihan]

I have never agreed that there is an optimum bias point for any type of output device (BJT or MOSFET).

Surely, the larger the bias current, the larger the output signal can be before the output stage changes from A to AB. Since we listen to signals which cross the zero point regularly, it is in our interest for the bias current to be as large as possible (so that we stay in class A more of the time)?

 

[SQLGuy]

What do you mean by accuracy? This is a critical question. After all, we don't design amplifiers for the listening pleasure of our oscilloscopes - although some seem to!

Ahh, this is why I asked Lumba what his goal was. But he never replied.

For me accuracy is exactly that, objective accuracy: output = input times some amplification factor. Nothing added, nothing removed, at least within the audio band. I don't care if the 2nd order distortion introduced by some single-ended 5W class A tube output stage adds a pleasant component that may sound better; that's not what I want.

And, for your "final arbiter", you make no mention of controlling the tests. Most "listeners" are categorically opposed to blind or double-blind tests. That's pure denial. They don't want to admit what they know to be true: that distortion analyzers and oscilloscopes ARE better at testing amplifiers than humans weighed down by prejudices and placebo effects.

 

[jan.didden]

I have never agreed that there is an optimum bias point for any type of output device (BJT or MOSFET).

Surely, the larger the bias current, the larger the output signal can be before the output stage changes from A to AB. Since we listen to signals which cross the zero point regularly, it is in our interest for the bias current to be as large as possible (so that we stay in class A more of the time)?

Well, at some point you ARE going to switch from A to AB (except full class A amp of course). So you may want to find the point where the takeover from one polarity device to the other polarity device is smoothest. That may be your optimum bias point.

So it's a tradeoff. You may stay longer in class A, where the switch-over occurs less but is more objectionable, or use an 'optimum' bias where the change-over occurs more often but is less objectionable. This obviously is different for a different amp topology and types of devices, as well as personal preference.

jd

 

[traderbam]

For me accuracy is exactly that, objective accuracy: output = input times some amplification factor. Nothing added, nothing removed, at least within the audio band. I don't care if the 2nd order distortion introduced by some single-ended 5W class A tube output stage adds a pleasant component that may sound better; that's not what I want.

I agree that a distortionless system will sound best. This is like saying a world without cruelty would be best. It's true but it's unhelpfully impractical. Given that distortion exists, it would be a mistake to assume that our ear-brain is equally sensitive to all distortions. That's why tube sound can be very persuasive in spite of gobs of THD.

And, for your "final arbiter", you make no mention of controlling the tests. Most "listeners" are categorically opposed to blind or double-blind tests. That's pure denial. They don't want to admit what they know to be true: that distortion analyzers and oscilloscopes ARE better at testing amplifiers than humans weighed down by prejudices and placebo effects.

Sure. Always listen carefully. But don't make the mistake of thinking the human brain-ear is an illegitimate witness. That would be silly since this is the witness we are designing for.

T&M equipment is arguably better at making a few, repeatable measurements with a certain accuracy. Usually. Let's not underestimate the ear-brain's sophistication for processing sound waves in a real-time manner. T&M equipment and simulator measurements only capture little slices of the big picture in this respect. If you become seduced by your T&M equipment you'll end up designing hifi that pleases your T&M equipment.

 

[ihan]

Hi jd,

Thanks for replying. I agree its a compromise:-

There is an optimum bias point if the input signal has a constant amplitude - it can be easily found with a distortion or audio analyser.

Since music is a series of summed sinewaves of continuously varying amplitude, frequency and phase, I think selecting an optimum bias is a difficult problem to solve.

My approach is to use as much bias as can be thermally tolerated (or as much as the user is willing to pay for in electricity bills).

 

[SQLGuy]

I think there is a key point of our disagreement here. It seems to me that you are saying that a device that sounds better is more accurate, in some way that cannot be measured by current test equipment. I believe that the test equipment is correct, and that amplifiers that distort more (and are less accurate) may still sound better to some people because the distortion components are euphonic.

Chorus and reverb undeniably reduce accuracy and add distortion of the audio signal, yet vocals with these effects added can sound better than "dry" vocals.

I want the measurable accuracy.

 

[jan.didden]

Hi jd,

Thanks for replying. I agree its a compromise:-

There is an optimum bias point if the input signal has a constant amplitude - it can be easily found with a distortion or audio analyser.

Since music is a series of summed sinewaves of continuously varying amplitude, frequency and phase, I think selecting an optimum bias is a difficult problem to solve.

My approach is to use as much bias as can be thermally tolerated (or as much as the user is willing to pay for in electricity bills).

It doesn't matter whether it is a sine wave or music. The important point is where the signal goes through the switchover point, and that is the same for sine waves or music or whatever.

jd

 

[Ultima Thule]

Vre 26 mV

that is not a question I could see.
You asked if it was conventional ClassB outputs switching off at quiescent or ClassAB passing 20 to 50mA.

Self sets up his Blameless by Vre not by Ire/Ib.
Re=0r1, 21.3mVre
Re=0r22, 23.1mVre
Re=0r33, 23.8mVre
Re=0r47, 27.4mVre

I'm not sure why he came up with that high value of 27.4mVre for Re=0r47.
Most commentators say it should be <26mVre and some say reducing as Tj rises.

(not a response to Andrew directly..)

just noticed the "proper" biasing of BJTs in the OP stage is finally up for talk...

sajti had a picture attached to his post #71 http://www.diyaudio.com/forums/soli...-audio-power-output-stages-4.html#post2015988 whereby wahab made some simulations with Iq 240 mA, I then asked in post #85 wahab to make a new simulation with Iq 52 mA on the OPS, unfortunately it passed by many readers attention back then that the suggestion of Iq of 52 mA was based on the 26 mV Vre, http://www.diyaudio.com/forums/soli...-audio-power-output-stages-5.html#post2017186.

Anyhow wahab's simulations showed the distortion worsened for the lower Iq of 52 mA, has anybody got any good simulation results with the "optimum" Vre on BJTs or maybe the "optimum" biasing doesn't work in simlations?

Cheers Michael

 

[ihan]

It doesn't matter whether it is a sine wave or music. The important point is where the signal goes through the switchover point, and that is the same for sine waves or music or whatever.

jd

My point is that the optimum bias setting (if you require minimum THD) is dependent upon the input signal amplitude.

A small input signal only requires a small bias current for the amp to remain in class A. A larger input signal requires a larger bias current to give a similar level of distortion.

If the bias current is made high enough, then for a larger percentage of time you will be listening to an amp in class A, and AB will only be entered into with large transients.

 

[ihan]

Hi Jan,

Let me try and highlight my point by posing a question:-

For example, if 100mA of bias current gives the lowest distortion with an output signal level equivalent to 10W of power, does that mean that we should definitely not bias the amp above 100mA, even though doing so would increase the output signal range that the amp would stay in class A?

An interesting question - wouldn't you say?

Regards,
Ian

 

[tiefbassuebertr]

Hi jd,

Thanks for replying. I agree its a compromise:-

There is an optimum bias point if the input signal has a constant amplitude - it can be easily found with a distortion or audio analyser.

Since music is a series of summed sinewaves of continuously varying amplitude, frequency and phase, I think selecting an optimum bias is a difficult problem to solve.

My approach is to use as much bias as can be thermally tolerated (or as much as the user is willing to pay for in electricity bills).

This approach was the same like that from Linn's amplifier models "LK-2", "LK-280" and "Klout" (from Glasgow/Scotland). All this models runs between 100mA and 120mA idle current through the output power stage by +/-41V regulated supply voltage (by the use of BjT and no power MOSFET output devices). This means 16-17 watts loss power for each channel and relatively warm heatsinks without output power. By low levels (<200 mA load current) they run in pure class A and sounds better than lower idle current values. at higher output current levels through the load the switchover is more clearly audible than by the optimum of idle current value (see janneman post #388)

Typically, the quiescent current is only by MOSFETs in Class AB applications in this aera and for power BJT's was in most cases select current values between 20 and 50mA (where the take-over from one polarity device to the other polarity device is smoothest - as janneman mentioned at post #388)
please read in this case also post #292 and #319 of me (my approach for a royal way), also #310 from Eva (all this mentioned posts from this thread)

 

[Telstar]

The problems with global feedback are serious enough even if properly used, noticed by many audiophiles, those who use their ears.
The time delay is a huge and fundamentally insurmountable problem. There´s every reason to doubt the accuracy of time delay and phase correction techniques.

On this point, I agree with Lumba. I do not have seen any proof that would state otherwise (i.e. that GLOBAL nfb do NOT cause smearing -not just delay- in the time domain).

Global feedback lowers the measured distortion by transforming (multiplying) harmonics into high frequency dirt, safely outside the measurable range, thus it`s essentially about distortion redistribution.
It also inevitably introduces new distortions. The result is an adverse harmonic spectrum, tones deprived from their harmonic content, a tangle of phase relationships, a clean but unpleasurable sound.

This may or may not be true. But for me it's a fact that power amplifier with not tiny GNFB sound harsh in the HF.

 

[Telstar]

Also, from what I understand, Baxandall has shown how nfb whether local or global, increases higher order harmonics in non-linear systems. So what we see from the application of nfb is lower THD which shows up particularly well in terms of reduced low order harmonics, but we also introduce more high order harmonics. One could describe this as a redistribution from low order to high order. I'm not saying this is a bad thing, but isn't this correct and also consistent with what Lumba posted?

I am. It IS a bad thing. Harmonics above the third are innatural.

I'm thinking that nfb is the greatest thing since sliced bread for reducing distortion from non-linear amplifier elements which all of our amplifiers have inside of them.

Anybody is entiled his own opinion.

 

[Telstar]

Stuff well above 20khz does not matter, it's above the frequency range that the tissues in our ear can pick up, and outside the frequency range that tweeters can play efficiently. Class D takes advantage of that.

This is exacly why most solid state amps sounds horrible in the high frequency. And class D ones even worse.

 

[Telstar]

He's incorrect in attributing this problem to the feedback itself instead to the BAD USE of feedback.

I attribute it to the use of GLOBAL NFB or to NFB in the output stage.

Timing problems are out of worries from at least a 40 years, as planar devices become largely available on market.

If you mean that time problems have been underestimated for 49 years, then we agree

Lumba Ogir, as many others, don't know - or forget to know - that if most higher frequency we can ear is about 20 kHz, then we can't ear time slices shorter than of the corresponding period.

See my other post, and also tiefbassuebertr's who explained better than me.
Even Self measured THD higher than 20khz

Actually appear that real abilities in detecting time slices of human ears are at most limited to the frequency band where its sensitivity peaks - 2-3 kHz, roughly equivalent to a 0.5 ms... when typical throughtput timing of transiting signal in not exceptional audio electronics are AT LEAST 500 times shorter! Time is not anymore a problem in audio electronic, either analogical or digital (the so called "jitter problem", in the form as often publicized in audio reviews, is a not problem at all...).
It should be more consistent if Lumba don't missed again a key factor for judging this event: ITS MAGNITUDE. If its true that NFB spread up the spectral population of distortion, its also true that the total amount distortion is WELL lowered (and often lowered well deep into noise floor).

And you miss one important point: the higher the order of the distortion, the higher the sensitivity of the human ear.
So, while on paper a 0,01% amplifier with 0,01% of all orders up to 6th distortion LOOKS better than one with 0,1% mostly 2nd harmonic, i wont bet which one sounds more natural , especially in the HF.

But is not this the real missed point. That really matter when distortion sink below noise floor is its ability to change meaningfully the "color" of the noise floor itself: if this remain in his composition substantially "white" not with pure measuring tones but with the true complex riddles of effective audio signal, then we can consider distortion essentially wiped out from the acoustical scene.

This I agree. But where is the noise floor? It's well below the typical instrumentation avaiable to most hobbists. I think it is below 0,01%.

This is the true reason what, sometime, manage some strumentally deficitary amplifiers to "sound better than" of other REALLY (I well underline this word: REALLY) instrumentally better amplifiers: simply when mixed down to listening, their features of noise floor, spectrum of distortion manage to form with music a "syndrome" that hear more "pleasurable" (essentially because SIMPLER THAN, and thus less fatiguing than original audio signal)[/quote]

I see this a bit differently.
The original event is never fatiguing nor harsh or unpleasant to the ears. Reproduced audio signals very often are. But dont worry the culprit most often is not the amplifier but the AD and D/A process.
Tubes and similar built amplifiers add pleasant distortion that makes it more pleasurable. an attempt to fix what is flawed from the start.
In the ideal world the source would be as "blameless" as current technology can make it, and the amplifier done with extremely low distortion (of all kind not just THD) as well, with high slew rate and without time smearing.

The point about PSU is extremely good and you know how much we agree on this. why? Because it is the main cause of HF distortions