darkfenriz said:
Yes, I guess that is correct. The linear distortion can modify the working point of a non-linear device in an amp and as such modify the non-linear components in the output. But I would say there is a 50-50 chance that the output improves, that the total non-linearities DECREASE. We are a long way from having a clear view on THAT, and I don't think this point has been addressed in the discussion, nor was this point made by anyone unless I completely missed some points (which has happened before..).
On the other hand, putting a signal through a non-linear device will surely make the output more non-linear, and generally we can exactly compute how much and in what way.
My purpose in the last few posts is to try to find out what we all mean, because I have the feeling we are loosely discussion a subject but we all use different terms and meanings.
Jan Didden
Hi, Janneman,
You are confusing me
When I talked about the non linearity of transistor(s)
http://www.diyaudio.com/forums/showthread.php?postid=671853#post671853
you said that the feedback as a "black box" will fix it.
Or is it you are looking inside the "black box" now?
You are confusing me
When I talked about the non linearity of transistor(s) http://www.diyaudio.com/forums/showthread.php?postid=671853#post671853
you said that the feedback as a "black box" will fix it.
Or is it you are looking inside the "black box" now?
David,
No black box now, I was referring to a component (transistor, diode, FET) with a non-linear characteristic, ie a characteristic (like gain or interterminal capacitance) that changes under the influence of the signal, versus a linear component that doesn't change under the influence of a signal, ie the capacitance of a capacitor, the inductance of a coil or the resistance of a resistor.
A linear component, or a network of linear components like a xover networ, will not create harmonic or IM distortion, only linear distortion (non-linear freq and phase response).
Jan Didden
No black box now, I was referring to a component (transistor, diode, FET) with a non-linear characteristic, ie a characteristic (like gain or interterminal capacitance) that changes under the influence of the signal, versus a linear component that doesn't change under the influence of a signal, ie the capacitance of a capacitor, the inductance of a coil or the resistance of a resistor.
A linear component, or a network of linear components like a xover networ, will not create harmonic or IM distortion, only linear distortion (non-linear freq and phase response).
Jan Didden
Ah, I see. I agree with that. They will not be able to make IM, but feeding it back to transistors (non linear gain device) does.
I want to ask a question. What do you think of feedback system (input differential) that is built with non-linear devices (transistors), will it be able to do good job in managing non-linear distortions (like the so called speaker nonlinearities with all its components)? Or, they do the opposite, makes worsen things?
But please don't see it in the scope of "black box", because looking at this by comparing input and output (feedback theory) will always give right answer
IF transistor is a linear gain device, I would agree that feedback is the ultimate cure (this is funny
, like a member said, if transistor is already linear gain device, why we need feedback at all?)I'm not saying making non-feedback amp is the ultimate way. Making non-feedback amp is having the same difficulties with making feedback amp, because the transistor itself is inherently not linear. In front end(s), it is easier to make linear performance, like with heavy degeneration. But it will stuck at the end with output stage(s).
In the situation where none is giving perfect answer, which is better, making feedback or non feedback audio amplifier with (non-linear) transistor(s)?
Hi Barry,
Personally I would keep the noisy rectifiers/psu/line coupled transformer wires away from the pcb.
You do not need snubbers at the fuses when there is a large C either side.
I can assure you that the flash of a blowing fuse with this circuit is wire melting incandescence, and not a spark. (I'm sure Carlos would agree.)
It is also possible that you could parallel a pair of resistor holes with each fuse for 22 ohm then 0.22 ohm resistor fitment to facilitate initial start up testing and biasing; prior to finalising with fuse fitment.
(How will you do this if the outputs devices are all pcb mounted - and - bolted to the heatsink?)
You have the VAS and driver transistors shown as TO92 and not TO126 size, yet I do not know of any TO92 devices that could do this job.
Also the Vbe multiplier should be mounted on the heatsink for thermal feedback.
My suggestion is that everything from the VAS base onwards is mounted and hard wired on the heatsink, with all wires coming to a 4x TO92 pcb.
This should have a substantial central star earth. (I have redrawn the circuit to illustrate my layout, though nothing is drawn out.)
Each rail fuse should first supply rail capacitors and then star out from this separately to components plus heatsink; ie. do not have output device current through track resistance generating a fractional voltage that will modulate supply to the input stage.
My biggest worry with your existing layout is that your input socket is incapable of grounding interference on the screen of the input cable wrt amplifier input.
If you wish to have output devices and drivers pcb mounted then split the circuit and use two pcbs with separation at the VAS base and the positive driver base.
I trust you now understand what is meant by 'star' earth/grounding.
Keep the class-A power device physically closest to the VAS, and keep the inputs 5cm/2" away from output wires/Zobel etc.
Cheers ........... Graham.
Personally I would keep the noisy rectifiers/psu/line coupled transformer wires away from the pcb.
You do not need snubbers at the fuses when there is a large C either side.
I can assure you that the flash of a blowing fuse with this circuit is wire melting incandescence, and not a spark. (I'm sure Carlos would agree.)
It is also possible that you could parallel a pair of resistor holes with each fuse for 22 ohm then 0.22 ohm resistor fitment to facilitate initial start up testing and biasing; prior to finalising with fuse fitment.
(How will you do this if the outputs devices are all pcb mounted - and - bolted to the heatsink?)
You have the VAS and driver transistors shown as TO92 and not TO126 size, yet I do not know of any TO92 devices that could do this job.
Also the Vbe multiplier should be mounted on the heatsink for thermal feedback.
My suggestion is that everything from the VAS base onwards is mounted and hard wired on the heatsink, with all wires coming to a 4x TO92 pcb.
This should have a substantial central star earth. (I have redrawn the circuit to illustrate my layout, though nothing is drawn out.)
Each rail fuse should first supply rail capacitors and then star out from this separately to components plus heatsink; ie. do not have output device current through track resistance generating a fractional voltage that will modulate supply to the input stage.
My biggest worry with your existing layout is that your input socket is incapable of grounding interference on the screen of the input cable wrt amplifier input.
If you wish to have output devices and drivers pcb mounted then split the circuit and use two pcbs with separation at the VAS base and the positive driver base.
I trust you now understand what is meant by 'star' earth/grounding.
Keep the class-A power device physically closest to the VAS, and keep the inputs 5cm/2" away from output wires/Zobel etc.
Cheers ........... Graham.
I think there is no question that any feedback system with time lag may generate nonlinear behavior. One simple example is oscilllation. The usual approach indeed considers the feedback system as a black box, and assumes that circuit configurations taking e.g. the Bode plot into account etc, will suffice to make the system stable.
My understanding of Graham's position is that
1- he believes that the black box approach does not suffice to explain the system behavior satisfactorily (the system does not conform to the model, i.e. is non ideal)
2- he believes that even if the black box approach was valid as a whole, that its equations only describe steady state behavior and that transients behave differently (the system misbehaves due to its intrinsic nature, even if ideal).
So, the questions are, is that right, and if yes, does it matter in practice?
I might add that my lack of technical background does not allow me substantial conclusions here ;-) ... I keep an open mind. I am especially suspicious about single formula models.
My understanding of Graham's position is that
1- he believes that the black box approach does not suffice to explain the system behavior satisfactorily (the system does not conform to the model, i.e. is non ideal)
2- he believes that even if the black box approach was valid as a whole, that its equations only describe steady state behavior and that transients behave differently (the system misbehaves due to its intrinsic nature, even if ideal).
So, the questions are, is that right, and if yes, does it matter in practice?
I might add that my lack of technical background does not allow me substantial conclusions here ;-) ... I keep an open mind. I am especially suspicious about single formula models.
oh yes, and by definition, feedback comes after the fact (the feed before the back). Therefore the error correction comes after the error. Therefore the initial error *must* remain uncorrected for. I see a logical necessity in favor of Graham's argument.
The question again is, magnitude and relevance.
The question again is, magnitude and relevance.
MBK said:
Sorry, in fact every oscilator needs a nonlinear section (unrelated with feedback) to operate.
What feedback ensures is that the closed loop transfer function contains a pole on the right plane (well, a conjugate pair), which implies an unbounded ramp-up of oscillation amplitude.
The nonlinear part may be implicit (e.g. power supply clipping) or explicit in the form of some gain control (lamps, thermistors or more elaborate gain controled stages are used for this purpose).
Sure enough, and this is accounted for by the frequency dependent expression of the closed loop transfer function. The result is the amount of correction is frequency dependent.
This may be understood intuitively. If the signal varies very slowly, the feedback delay is negligible and correction attains its maximum effect. If the signal is significantly fast with respect to the system delay (measured as phase shift), correction is less effective.
All this can be put very precisely in formal terms, and is to a large degree (among other things) what Control Systems Theory is about.
It is also true negative feedback cannot eliminate errors completely by its very nature (in cannonical topology) but can reduce them below significance.
Rodolfo
Hi Jan,
To me 'linearity' has always been 'continuous proportionality'.
That is a fundamental definition that was not written by me, and it is not wrong.
Amplifier amplitude linearity can be measured in time isolation to specify a good thd figure, and yet reproduction can still sound poor with dynamic waveforms because of the reactive delays associated with its internal stabilisation and NFB circuitry.
For those who don't deny its existence, this is where First Cycle Distortion observation will show on the simulator which amplifiers are capable of clean dynamic reproduction; like the JLH class-A is within its working capabilities.
In your Post#557 you write about me giving the impression ........
Your definition of a 'linear' circuit was been different to mine, that is why you have developed impressions different to what I was actually stating from my understanding of the word 'linear'.
What I do know however, is that many members have categorically stated that a series output choke (ie. cable too) does not develop an error voltage because it is a 'linear' component, which to me means a claim for continuous proportionality. I also know that I have been treated very badly when I say that a choke/cable does introduce a leading error voltage wrt amplifier output, even though I have gone out of my way in trying to explain the reasons for this arising.
So then , a series choke (or cable) does not transfer audio voltage output with 'continuous proportionality' when the load is complex, (it does with a resistor - but not with a loudspeaker !!! and as Darkfenriz concurs, this is because the group delay is not constant with frequency when the load is reactive). Also, the error voltage developed by the choke/cable fractionally series modulates the original amplifier voltage output.
I have not made any claims for harmonic generation either (!!!), merely that the high frequency nature of series output choke/cable error voltage has the greatest impact upon detail, characteristic and positional information carried by high harmonic frequencies.
If all of this really has been widely known for many years, then I wonder why the argument (and the inappropriate name calling/mischeivous postings by some !) ?
(Jan, to me your use of the term 'linear distortion' is completely meaningless. I'm not trying to be awkward, but maybe from the forgoing you can understand my point of view, as I now understand how you came to use this term ! Thus I am now having to translate your text in my head as I go along. No doubt you now view mine differently too !)
Your Post#561.
I did broach this, but if there is a difference then there is an error, and Andy has isolated 'something'.
Same hymn sheet ? Let's hope so, because this is only one of many distortions !
Cheers ........... Graham.
To me 'linearity' has always been 'continuous proportionality'.
That is a fundamental definition that was not written by me, and it is not wrong.
Amplifier amplitude linearity can be measured in time isolation to specify a good thd figure, and yet reproduction can still sound poor with dynamic waveforms because of the reactive delays associated with its internal stabilisation and NFB circuitry.
For those who don't deny its existence, this is where First Cycle Distortion observation will show on the simulator which amplifiers are capable of clean dynamic reproduction; like the JLH class-A is within its working capabilities.
In your Post#557 you write about me giving the impression ........
Your definition of a 'linear' circuit was been different to mine, that is why you have developed impressions different to what I was actually stating from my understanding of the word 'linear'.
What I do know however, is that many members have categorically stated that a series output choke (ie. cable too) does not develop an error voltage because it is a 'linear' component, which to me means a claim for continuous proportionality. I also know that I have been treated very badly when I say that a choke/cable does introduce a leading error voltage wrt amplifier output, even though I have gone out of my way in trying to explain the reasons for this arising.
So then , a series choke (or cable) does not transfer audio voltage output with 'continuous proportionality' when the load is complex, (it does with a resistor - but not with a loudspeaker !!! and as Darkfenriz concurs, this is because the group delay is not constant with frequency when the load is reactive). Also, the error voltage developed by the choke/cable fractionally series modulates the original amplifier voltage output.
I have not made any claims for harmonic generation either (!!!), merely that the high frequency nature of series output choke/cable error voltage has the greatest impact upon detail, characteristic and positional information carried by high harmonic frequencies.
If all of this really has been widely known for many years, then I wonder why the argument (and the inappropriate name calling/mischeivous postings by some !) ?
(Jan, to me your use of the term 'linear distortion' is completely meaningless. I'm not trying to be awkward, but maybe from the forgoing you can understand my point of view, as I now understand how you came to use this term ! Thus I am now having to translate your text in my head as I go along. No doubt you now view mine differently too !)
Your Post#561.
I did broach this, but if there is a difference then there is an error, and Andy has isolated 'something'.
Same hymn sheet ? Let's hope so, because this is only one of many distortions !
Cheers ........... Graham.
lumanauw said:
Ah, I see. I agree with that. They will not be able to make IM, but feeding it back to transistors (non linear gain device) does.
I want to ask a question. What do you think of feedback system (input differential) that is built with non-linear devices (transistors), will it be able to do good job in managing non-linear distortions (like the so called speaker nonlinearities with all its components)? Or, they do the opposite, makes worsen things?
But please don't see it in the scope of "black box", because looking at this by comparing input and output (feedback theory) will always give right answer
IF transistor is a linear gain device, I would agree that feedback is the ultimate cure (this is funny
I'm not saying making non-feedback amp is the ultimate way. Making non-feedback amp is having the same difficulties with making feedback amp, because the transistor itself is inherently not linear. In front end(s), it is easier to make linear performance, like with heavy degeneration. But it will stuck at the end with output stage(s).
In the situation where none is giving perfect answer, which is better, making feedback or non feedback audio amplifier with (non-linear) transistor(s)?
David,
There is enough here for many pages, I don't think I have the time now. Some comments:
"They will not be able to make IM, but feeding it back to transistors (non linear gain device) does. "
Who is they, what do they feed back?
"In the situation where none is giving perfect answer, which is better, making feedback or non feedback audio amplifier with (non-linear) transistor(s)? "
Please define 'better', then I'll give you a meaningfull answer. There are many ways to make an amp, with a whole spectrum between o fb and heavy fb or local fb etc. They all give good results, if competently designed, with differences either not audible or only with very specific tests and very specific test signals. So, in a sense, there is no 'best' way. What is the 'best' way to make a car?
"But please don't see it in the scope of "black box", because looking at this by comparing input and output (feedback theory) will always give right answer
"Why not? This is important. If two black boxes give the same results, who cares what is inside? Black boxes are a great way to analyse and synthesise circuits, and it is not just a gimmick, it WORKS.
Jan Didden
Graham Maynard said:
Agreed, fully
Graham Maynard said:
Graham, not to split hairs, but: yes, I agree, in the sense that there are freq and phase response errors. What you call FCD is just the normal, fully known and explainable reaction of a complex circuit to a waveform with a very wideband initial portion. Some amps cannot cope with that high bandwidt portion and they should not be blamed for it, because it is not an audio signal. Some amps, like the ones you mentioned, apparently can. Put the input through a suitable input filter and puff, gone is the FCD. If you don't agree that the term linear distortion is as I postulated, then there is NO distortion whatsoever. Unless, again, you have an amp that gets non-linear with a signal never foreseen for it like you suddenly starting sine waves. It's like building a plane and then blaming it because it cannot cruise under water.
But, my 'definition' of linear distortion is one that is often used in textbooks, to contrast it with non-linear distortion as I wrote above. So, that's where my understanding of it comes from.
Graham Maynard said:
Yes of course, and we all know why that is. THD measurements with a single freq is not the same as exercising the amp with a music signal. What's new here? I don't see what FCD (which, if it exists, is ONLY freq and phase response deviations, linear distortion), has to do with this.
Jan Didden
janneman said:
Watch out Jan !!!! What our dear Graham is implying, is **perfect** proportionality, i.e. the input waveform exactly passed to the output only scaled.
Of course you know a system may be linear in the strict sense and yet not comply with the above because of frequency dependent amplitude/phase deviations.
Graham, I do belive you are a passionate and dedicated person, with most probably a proven track record of good designs under your belt (though I never had the honor to listen to one).
But please PLEASE, admit you lean on the stubborn side, this said with due respect.
This stubborness has infuriated me on more than one ocassion, and most probably is at the roots of less benign posts.
Please, PLEASE acknowledge there are individuals here who know what they are talking about, and deserve respect as much as you. Not that we should be waving resumes and academic credentials, only we have studied what it takes, and live professionaly from this.
Again, let me stress I trust both your experience and well meant convictions, an let me - I dare to say in the name of others - require you to acknowledge we are also well meant when we raise our voices to contradict some of your oppinions based on what we know as professionals. Doing otherwise on your part, can only end in keeping you stuck in error and us giving up and looking elsewhere.
Well I had to say it sooner or later.
Rodolfo
lumanauw said:
I want to ask a question. What do you think of feedback system (input differential) that is built with non-linear devices (transistors), will it be able to do good job in managing non-linear distortions (like the so called speaker nonlinearities with all its components)? Or, they do the opposite, makes worsen things?
But please don't see it in the scope of "black box", because looking at this by comparing input and output (feedback theory) will always give right answer
I'm not Jan, but
Your answer is in Hiraga's article you've posted at the beggining of this thread.
Well designed amps are able to whistand inverse signals (from output to input) without distorting, ill designed ones don't.
By sheer coincidence, tha best one in this regard had a very low output impedance power stage and lots of feedback...
Same as above, in my opinion.
Hi MBK,
My understanding is that it as wrong to let an amplifier develop a steady sinewave before analysing its amplitude linearity for thd in isolation as it is to observe series output choke induced error with a steady sinewave.
The amplifier may still be a black box, but it should instead be monitored (simulated) for thd generation during a suddenly starting 10kHz first cycle.
Of course many have professed that I lack an understanding of everything related to thd or a suddenly starting sinewave etc., also that no such wave exists, but all they actually cover are obvious aspects that were once my own stepping stones whilst studying the never ending stream of first cycles that make 'music'.
So why not separately use sine and squarewave testing ? Because steady sinewaves do not test for dynamic response capabilities, whilst squarewaves drive beyond amplifier linearity throughout their peak to peak range.
However a suddenly starting sinewave already combines both sine plus square characteristics yet without driving the amplifier into a peak to peak non-linearity that proves nothing. Also an amplifier that shows itself better capable of coping with a suddenly starting 10kHz sine that cannot exist in real life is also going to be more linear when coping with the dynamically changing cycles that do.
It is actually possible to optimise stabilisation component values ( and some feedforward arrangements ) to improve the high audio frequency thd figure (time delayed error subtraction), but the increased distortion that will have arisen on first cycle linearity as a direct result would have passed into unrecorded history long before the normal thd measurement could accurately be made.
Many constructors continue to be amazed by JLH class-A amplifier transparency, and that is because its distortion is as low during a suddenly starting first amplification cycle ( wide activity bandwidth not just input bandwidth) as it is with any subsequent; it has no stabilisation components causing signal path delay, plus a direct feedback loop that is often described as being current feedback that gives it a mostly resistive output characteristic at higher frequencies.
Many solid-state amplifier designers would get a surprise if they set thd performance aside for just a moment and studied the first cycle (dynamic linearity) behaviour and output characteristics of their own creations.
Hi Rodolfo,
Dedicated - yes;
Stubborn - well I won't back out if I can see an end result for a resolvable problem, which equates to dedication.
Passionate - not quite kissy kissy.
Of course you have always been well intentioned, but some earlier posters were not.
Hi Jan,
I had the above already written for MBK before you posted !
You too are attacking my suggestion that FCD studies actually *do* have something to offer.
Maybe this has something to do with our deep founded differences in relation to 'linear distortion' ?
If by taking a different position to improve one's view, or by applying a different methodology to acheive results that no other test can reveal, then is it wrong to use that test ?
No; as long as the results are competently reviewed within the range of already known to be fundamentally correct and established findings. (Of which input filtering is one that I am well aware of.)
After-all, there are already equivalents of different methodologies in the arbiter if all theory - calculus itself !
If FCD "is just the normal" then why not study it ?
I suggest that denying the use of first cycle observation is to deny insight into amplifier circuit operation.
Cheers ........... Graham.
My understanding is that it as wrong to let an amplifier develop a steady sinewave before analysing its amplitude linearity for thd in isolation as it is to observe series output choke induced error with a steady sinewave.
The amplifier may still be a black box, but it should instead be monitored (simulated) for thd generation during a suddenly starting 10kHz first cycle.
Of course many have professed that I lack an understanding of everything related to thd or a suddenly starting sinewave etc., also that no such wave exists, but all they actually cover are obvious aspects that were once my own stepping stones whilst studying the never ending stream of first cycles that make 'music'.
So why not separately use sine and squarewave testing ? Because steady sinewaves do not test for dynamic response capabilities, whilst squarewaves drive beyond amplifier linearity throughout their peak to peak range.
However a suddenly starting sinewave already combines both sine plus square characteristics yet without driving the amplifier into a peak to peak non-linearity that proves nothing. Also an amplifier that shows itself better capable of coping with a suddenly starting 10kHz sine that cannot exist in real life is also going to be more linear when coping with the dynamically changing cycles that do.
It is actually possible to optimise stabilisation component values ( and some feedforward arrangements ) to improve the high audio frequency thd figure (time delayed error subtraction), but the increased distortion that will have arisen on first cycle linearity as a direct result would have passed into unrecorded history long before the normal thd measurement could accurately be made.
Many constructors continue to be amazed by JLH class-A amplifier transparency, and that is because its distortion is as low during a suddenly starting first amplification cycle ( wide activity bandwidth not just input bandwidth) as it is with any subsequent; it has no stabilisation components causing signal path delay, plus a direct feedback loop that is often described as being current feedback that gives it a mostly resistive output characteristic at higher frequencies.
Many solid-state amplifier designers would get a surprise if they set thd performance aside for just a moment and studied the first cycle (dynamic linearity) behaviour and output characteristics of their own creations.
Hi Rodolfo,
Dedicated - yes;
Stubborn - well I won't back out if I can see an end result for a resolvable problem, which equates to dedication.
Passionate - not quite kissy kissy.
Of course you have always been well intentioned, but some earlier posters were not.
Hi Jan,
I had the above already written for MBK before you posted !
You too are attacking my suggestion that FCD studies actually *do* have something to offer.
Maybe this has something to do with our deep founded differences in relation to 'linear distortion' ?
If by taking a different position to improve one's view, or by applying a different methodology to acheive results that no other test can reveal, then is it wrong to use that test ?
No; as long as the results are competently reviewed within the range of already known to be fundamentally correct and established findings. (Of which input filtering is one that I am well aware of.)
After-all, there are already equivalents of different methodologies in the arbiter if all theory - calculus itself !
If FCD "is just the normal" then why not study it ?
I suggest that denying the use of first cycle observation is to deny insight into amplifier circuit operation.
Cheers ........... Graham.
Graham,
I won't enter into technical discussion here because I'm always short of time, but it seems to me that first cycle distortion - perhaps pulse distortion by another name - is a perfectly valid way to address the seminal problems of audio amplifiers.
Of course, you must realise that anything unusual and seemingly wacky will always be challenged and derided by the cognoscenti. This is merely a rite of passage, and the challenges must be met for obvious reasons. My country, Australia, is quite anti-intellectual, but it's a great test and one you might warm to. Once convinced however, your greatest deriders often become your biggest fans, and I see this process as healthy. Who knows, you might become the next Arthur Bailey, or JLH!!
Now, can you tell me how the PNP output devices in your GEM amplifier always remain on? I can't see it; the NPNs, yes, but not the PNPs.
Cheers,
Hugh
I won't enter into technical discussion here because I'm always short of time, but it seems to me that first cycle distortion - perhaps pulse distortion by another name - is a perfectly valid way to address the seminal problems of audio amplifiers.
Of course, you must realise that anything unusual and seemingly wacky will always be challenged and derided by the cognoscenti. This is merely a rite of passage, and the challenges must be met for obvious reasons. My country, Australia, is quite anti-intellectual, but it's a great test and one you might warm to. Once convinced however, your greatest deriders often become your biggest fans, and I see this process as healthy. Who knows, you might become the next Arthur Bailey, or JLH!!
Now, can you tell me how the PNP output devices in your GEM amplifier always remain on? I can't see it; the NPNs, yes, but not the PNPs.
Cheers,
Hugh
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.