Nico, that specific complementary bipolar pair (MJ15003-4) is one of the worst for audio that I have ever used. I tried to use them with the Gale 250W amp, back in 1976, and I never could get them to sound as good as an Audio Research tube power amp with Stax electrostatic headphones. About 1980, ring emitter transistors were introduced and they are about 25 times faster and many times more linear. Today, we use ring emitter based transistors for serious bipolar output designs, and I drive them with a pair of complementary power mos fets.
Personally I would use mosfets for the output, if they would hold up in my circuits.
What I was trying to point out to Bob and anyone else, is that feedback can always lower distortion to virtually unmeasurable levels, if you use enough of it. That is how Halcro does it, and Bob's circuits use a fair amount as well. Many designers, including Charles Hansen of Ayre, try to make the most linear amps possible without resorting to global negative feedback, if possible. We sometimes have to live with a small amount of lower order distortion at listening levels to get the most natural sound possible. I do not know exactly why this is so, but open loop bandwidth might have something to do with it.
I simply wanted to point this out, i.e. that most professional designers know how to achieve low distortion amps, but not all amps made this way sound the same or even as good as other amps designed without ultra low distortion as the most important goal.
Personally I would use mosfets for the output, if they would hold up in my circuits.
What I was trying to point out to Bob and anyone else, is that feedback can always lower distortion to virtually unmeasurable levels, if you use enough of it. That is how Halcro does it, and Bob's circuits use a fair amount as well. Many designers, including Charles Hansen of Ayre, try to make the most linear amps possible without resorting to global negative feedback, if possible. We sometimes have to live with a small amount of lower order distortion at listening levels to get the most natural sound possible. I do not know exactly why this is so, but open loop bandwidth might have something to do with it.
I simply wanted to point this out, i.e. that most professional designers know how to achieve low distortion amps, but not all amps made this way sound the same or even as good as other amps designed without ultra low distortion as the most important goal.
john curl said:
John,
I have been using these transistors in the design simply because I have a tray full. The 2SK1058/2SJ162 is probably also not the best available, but both are more or less in the same cost ball park similar current and voltage characteristics. So my aim was to see if there were any obvious audible difference. If I used different topologies, etc. I could conclude nothing.
My point John, and reading between your lines as well, an amplifier sounds like its designer. Every professional designer's goal is to produce the "best" amplifier s/he can.
Kindest regards
Nico
Let us say, these MJ devices are not my 'reference' transistors, and they would tend to give transistor output stages a 'bad rap' as far as audio quality is concerned. It is best to listen as well as measure, to establish what works and what is merely OK.
I also had high hopes for the MJ15003,4 devices. After all, I spent 6 months designing an amplifier with them, and they were my only choice for making my design target. I did find that and earlier part pair, the 2N5884,6 series, worked better, giving twice the slew rate and 1/2 the distortion with the same drive stage, but voltage limitations allowed only 100W out, instead of 250W, which was the Gale design target. They sounded better too, for some reason.
I also had high hopes for the MJ15003,4 devices. After all, I spent 6 months designing an amplifier with them, and they were my only choice for making my design target. I did find that and earlier part pair, the 2N5884,6 series, worked better, giving twice the slew rate and 1/2 the distortion with the same drive stage, but voltage limitations allowed only 100W out, instead of 250W, which was the Gale design target. They sounded better too, for some reason.
Nelson Pass said:
Right on, Nelson! A sane voice in the silly five-second/five-hour/whatever ABX-demo wilderness.
Audible differences between different components might seem subtle on short listen and with limited familiarity, but can *grate* quite sufficiently to render a component unlistenable in the longer term.
A first comment on the difference in sound between devices:
I think it is important to make the difference between accuracy of reproduction and "better sounding".
To judge the accuracy we need to compare with a reference and this one is the life performance. Like S. Linkwitz is saying, people not used to listening to life performance but more used to recorded music are often biased in their judgment.
Wine tasting is different because there is no ultimate reference. We should not, in my opinion, experience audio like wine testing (except for pleasure of course).
Understanding where the differences are coming from is another story and is the real quest.
The remark from M. Curl on feedback is of course interesting. The paper from Boyk and Sussman is addressing this(http://www.its.caltech.edu/~musiclab/feedback-paper-acrobat.pdf). They try to understand using numerical analysis the distortion pattern created by feedback on the different stages existing in an amp ( push pull, common emitter, differential) and this for tube bjt and fet.
They demonstrate the intermodulation pattern created by feedback in a sort of regenerative way but their conclusions are vague.
Second, I still have question on the proper biasing of a push pull output stage.
I read the paper of Oliver on optimal biasing. It is not easy to read because of inconsistancies in notation and there are some typographic errors in my opinion.
He analysed the output impedance ( open loop) of a push pull amplifier and its variation with load current and biasing.
He used total variables ( exponential relation between i and v)and had to solve a set of transcendental equations ( including logs). The only way is via numerical intergration.
His conclusions are:
let 'go' be the transconductance at Io the bias current ( it is gm at Io)
let Rc be the resistance in the emitter leg and let R be the resistance of the loop including one transistor, the base resistor and the emitter. For high beta, R=~Rc
Then
if current i >> Io Rout tends to R =~Rc
if current i drops then Rout rises until i=4Io to Rc + 0.22/go
From there it depends on goRc
when i<<Io
if goRc = 0.5 then Rout = Rc + 0.22/go
if goRc = 1 then Rout = Rc
if goRc = 2 then Rout = Rc - 0.5/go
The rule of thumb is keep goRc between 0.5 and 1 for minimal Rout variation which is the cause of the distortion a low signal level.
let Delta Rout = (Rout-Rc) ;it is the variation in output resistance
We know that go= Io(mA)/26 then
goRc = 0.5 implies RcIo = 13 mV and Delta Rout = 5/Io
goRc = 1 implies RcIo = 26mV Delta Rout = 0 but peak
of 5/Io at i=4 Io
goRc = 2 implies RcIo = 52mV Delta Rout = -13/Io
1° To apply this rule: the current must be very stable with T. Should the Motorola thermapak BJT not be perfect?
2° To minimize Delta it looks like it is better to use a large Io but then the window of Rout variation with current is wider. Rout starts to vary with current at i/Io<100. Is this not the explanation of the use of many transistors in // ? Then a lower Io per transistor can be used reducing the window but also a low variation of Rout results because they are in // and Io total is large
3° For those who have red the Oliver paper, he rejects the use of this rule of thumb because the voltage drop on Rc is to low with respect to Vbe variation with T. It would require a very well stabilized Io. He proposes a solution with diodes and claims that the distortion is lower. What do you think?
Jean - Pierre (JPV)
I think it is important to make the difference between accuracy of reproduction and "better sounding".
To judge the accuracy we need to compare with a reference and this one is the life performance. Like S. Linkwitz is saying, people not used to listening to life performance but more used to recorded music are often biased in their judgment.
Wine tasting is different because there is no ultimate reference. We should not, in my opinion, experience audio like wine testing (except for pleasure of course).
Understanding where the differences are coming from is another story and is the real quest.
The remark from M. Curl on feedback is of course interesting. The paper from Boyk and Sussman is addressing this(http://www.its.caltech.edu/~musiclab/feedback-paper-acrobat.pdf). They try to understand using numerical analysis the distortion pattern created by feedback on the different stages existing in an amp ( push pull, common emitter, differential) and this for tube bjt and fet.
They demonstrate the intermodulation pattern created by feedback in a sort of regenerative way but their conclusions are vague.
Second, I still have question on the proper biasing of a push pull output stage.
I read the paper of Oliver on optimal biasing. It is not easy to read because of inconsistancies in notation and there are some typographic errors in my opinion.
He analysed the output impedance ( open loop) of a push pull amplifier and its variation with load current and biasing.
He used total variables ( exponential relation between i and v)and had to solve a set of transcendental equations ( including logs). The only way is via numerical intergration.
His conclusions are:
let 'go' be the transconductance at Io the bias current ( it is gm at Io)
let Rc be the resistance in the emitter leg and let R be the resistance of the loop including one transistor, the base resistor and the emitter. For high beta, R=~Rc
Then
if current i >> Io Rout tends to R =~Rc
if current i drops then Rout rises until i=4Io to Rc + 0.22/go
From there it depends on goRc
when i<<Io
if goRc = 0.5 then Rout = Rc + 0.22/go
if goRc = 1 then Rout = Rc
if goRc = 2 then Rout = Rc - 0.5/go
The rule of thumb is keep goRc between 0.5 and 1 for minimal Rout variation which is the cause of the distortion a low signal level.
let Delta Rout = (Rout-Rc) ;it is the variation in output resistance
We know that go= Io(mA)/26 then
goRc = 0.5 implies RcIo = 13 mV and Delta Rout = 5/Io
goRc = 1 implies RcIo = 26mV Delta Rout = 0 but peak
of 5/Io at i=4 Io
goRc = 2 implies RcIo = 52mV Delta Rout = -13/Io
1° To apply this rule: the current must be very stable with T. Should the Motorola thermapak BJT not be perfect?
2° To minimize Delta it looks like it is better to use a large Io but then the window of Rout variation with current is wider. Rout starts to vary with current at i/Io<100. Is this not the explanation of the use of many transistors in // ? Then a lower Io per transistor can be used reducing the window but also a low variation of Rout results because they are in // and Io total is large
3° For those who have red the Oliver paper, he rejects the use of this rule of thumb because the voltage drop on Rc is to low with respect to Vbe variation with T. It would require a very well stabilized Io. He proposes a solution with diodes and claims that the distortion is lower. What do you think?
Jean - Pierre (JPV)
Nelson Pass said:The problem with these sorts of comparisons is that they
are not usually optimized to throw audible differences into
sharp relief. It is important to have lived with a reference
system for a period of time and then toss in a variable.
I have found that, as with wine, the differences are most
apparent when you step down in quality from a familiar
reference.
If this can be done in a way that ensures the listeners aren't aware of what they're listening to, then it's valid.
Otherwise, I fear that all people will hear are their own prejudices.
Regards,
Suzy
Nelson Pass said:
If by this you mean "objectively measures better" versus
"sounds more like being there", then you are right - it is
important to make the distinction.
Perhaps to many a "Distinction without a difference?", but I think I find myself agreeing with your approach.
Cheers!
Russ
Nelson Pass said:
If by this you mean "objectively measures better" versus
"sounds more like being there", then you are right - it is
important to make the distinction.
Sorry, and with due respect.
This line of argument never fails to mistify me - understanding it is not marketing in disguise - which of course I asume is not the case.
It is simply not acceptable to admit the existence of "unknown" or "unmeasurable" effects placing audio more or less at par with paranormal phenomena or alien abductions.
Neither will I push a naive approach whereby if an amplifier measures better than certain objective limits, then it "must be all right" no matter what others say. This is neither conductive to improvements for that matter.
But what I won't buy is, given a measurement setup including all the signal chain up to the speakers and listening environtment, and in that setup one verifies the acoustic presure fields match accurately the soruce within reasonable tolerances as to what the ear can discriminate, that then someone else may come with a different setup which in principle is less objectively accurate yet is defended as more "close to reality".
This is not black magic, this is not quantum weirdness, this fairly well understood macroscopic physics. If subjective appreciation does not match objective measurement, then something is wrong and can be found, either in the listener or in the complete test setup.
Rodolfo
Upupa Epops said:Rodolfo, if your customers are deaf and they still need " juice sound ", is hardly persuade them, that is better to listen " objectively measured better "... It is probably dilema, in front are standing every designers sooner or later...Try to think, which version he select...
That's why I placed the caveat in the second line - I asume we are at least in this forum and thread - doing away with unconfessed marketing strategies.
Of course I understand marketing considerations and the constraint of being succesfull in business may dictate departures from what could be acknowledged unremitting adherence to accuracy.
Rodolfo
ingrast said:
I've been waiting so long for that to happen i decided to become one myself. If you happen to bump into something green or blueish, preferable on the "Eve" side, hand out some directions.
The discussion on measurement before tasting the pudding in contrast to hearing and verifying with numbers seems like an endless clash between two opposing elements.
Funny thing, the whole MOSFET era started with both technical arguments on how much easier it is to drive MOSFETs, AND with the rationale that they sound better because of their softclipping nature and the "tube"-like blah blah.
Seems to me like the parties on both side of the fence were wrong.
Sounds like it isn't that favorable to skip the drivers on MOSFET output stages and without an additional EC circuit a class AB MOSFET output stage can not compete with BJTs, according to the statements of both the ones with ears and tools.
MOSFET power amps have come from all over the place.
Haflers from the US, Perreaux from Australia, Musical Fidelity from GB, Scylla and Sphinx from where i live, AVM and Neumann from Germany, to name just a few.
Thomas Hohne, the guy who set up the Neumann MOSFET amplifier brand in Germany also markets the Aaron label. Long time ago Mr Hohne personally tried to persuade me to buy one of his class AB Neumann amps.
Peculiar thing, the MJ15003/4 output stage Aaron amps of the same guy "sounded" far more acceptable than the Neumann horror.
Same goes for every other class AB MOSFET name i mentioned above, in any circuit topology. Even a Conrad Johnson Evo2000 tube/MOSFET hybrid is not everyone's flavor.
Which is not the case for high bias amps like the big MF and Perreaux type numbers, the full class A opamp input MOSFET amplifier of AVM and a full bias tube input Sphinx Project 16.
Let's not forget to mention Mr P's products.
Both Mr Curl, Mr Cordell, AND Nelson Pass seem to have their facts more straight that many others, imho.
Mr Ras must be one heck of a designer.
jacco vermeulen said:
These should not be opposing elements. That they look so can be attributed to at least a couple of facts.
Manufacturers need strategies to focus on in order to make themselves visible, and a common strategy is to pull away "sonics" features from hard objective data.
Audio is a particularly fertile ground for this, given the infinite ability of human beings for self-deception, which leads to the second fact.
I confess to find my amplifiers better than every other one, for I "know" they are designed with the strategy I consider best, and measure very well. Of course the same must happen with other designers and their cherished works, no matter how different in design and performance they could be.
One can rationalize on particular design strategies based on known facts like gm vs. current or clipping behavior, and it is OK as guides for implementation, but the proof is in the pudding and as Bob has often quoted, the devil is in the details.
The bottom line is that given equal listening setups, including speakers and environment, a system that succeeds in reproducing accurately a sound field as originaly aquired, cannot be judged inferior to other whose objective prformance measured this way falls below.
If that happens, it should not be attributed to "unknown" or as yet "unmeasurable" factors, something is wrong with the listener or the setup or both, and it can be traced and explained.
Rodolfo
