I observed (maybe wrong), that a power amp with single differential gives more focus in sound reproduction. It is not about nice to listen to, but more focus.
I bought Douglas Self's book and asked him why he didnt write at all about power amp with dual differential (transistor pairs, 1pair npn and 1 pair pnp). He answered that he suspected that dual differential power amp is not as linear as single differential amp.
Is it true that single differential amp (like aleph) is better than dual differential amp (most of the pro-audio amp, like crown, crest, uses this dual differential).
What is really the different between building power amp with single differential and dual differential? Maybe both can produce music that is nice to listen to, but is single differential more focus (more linear, like Douglas Self said?)
I bought Douglas Self's book and asked him why he didnt write at all about power amp with dual differential (transistor pairs, 1pair npn and 1 pair pnp). He answered that he suspected that dual differential power amp is not as linear as single differential amp.
Is it true that single differential amp (like aleph) is better than dual differential amp (most of the pro-audio amp, like crown, crest, uses this dual differential).
What is really the different between building power amp with single differential and dual differential? Maybe both can produce music that is nice to listen to, but is single differential more focus (more linear, like Douglas Self said?)
lumanauw:
Which do you mean, dual differential, or complementary differential?
Example of dual differential circuit:
http://k-amps.8m.com/cgi-bin/i/PowerAmps/Semicond/kaneda_mosfet.jpg
Example of complementary differential circuit:
http://www.klausmobile.narod.ru/industrial/in_01f_r2_schdetail.gif
regards, jonathan carr
Which do you mean, dual differential, or complementary differential?
Example of dual differential circuit:
http://k-amps.8m.com/cgi-bin/i/PowerAmps/Semicond/kaneda_mosfet.jpg
Example of complementary differential circuit:
http://www.klausmobile.narod.ru/industrial/in_01f_r2_schdetail.gif
regards, jonathan carr
The first schematic (Kaneda) is single differential (TR1,2), with upper cascode (TR3,4). The second schematic (schdetail.gif) is dual differential (maybe I should say complementary differential) the upper is NPN (TR1,2), and lower is PNP(TR3,4).
Thanks for the example. This is exactly what my question is all about. Single differential (like kaneda) VS complementary differential (like schdetail.gif).
Thanks for the example. This is exactly what my question is all about. Single differential (like kaneda) VS complementary differential (like schdetail.gif).
> single differential gives more focus in sound reproduction.
This has to be an over-generalization. Either type can be good or bad.
The choice is mainly about the fact that a NPN-PNP design can give outputs at both rails, which affects design of the second stage. It may be better, it can be worse.
Anyway: The difference you describe probably has a lot to do with driver and output stage linearity, more than input topology.
> Douglas Self... he didnt write at all about power amp with dual differential (transistor pairs, 1pair npn and 1 pair pnp). He answered that he suspected that dual differential power amp is not as linear as single differential amp.
I'm not sure I agree that it is worse, but for the levels used in audio input stages it should be about the same.
If the effective input voltage, under feedback, rises from a milliVolt to over 20mV, some types of NPN-PNP complementary diff-pair can hold "linearity" a little further than a single pair. But operation in that range isn't the "very-good linearity" we expect in audio. It may be useful in comparators and other mainly non-linear devices which need a semi-linear range for smooth settling.
This has to be an over-generalization. Either type can be good or bad.
The choice is mainly about the fact that a NPN-PNP design can give outputs at both rails, which affects design of the second stage. It may be better, it can be worse.
Anyway: The difference you describe probably has a lot to do with driver and output stage linearity, more than input topology.
> Douglas Self... he didnt write at all about power amp with dual differential (transistor pairs, 1pair npn and 1 pair pnp). He answered that he suspected that dual differential power amp is not as linear as single differential amp.
I'm not sure I agree that it is worse, but for the levels used in audio input stages it should be about the same.
If the effective input voltage, under feedback, rises from a milliVolt to over 20mV, some types of NPN-PNP complementary diff-pair can hold "linearity" a little further than a single pair. But operation in that range isn't the "very-good linearity" we expect in audio. It may be useful in comparators and other mainly non-linear devices which need a semi-linear range for smooth settling.
It is true that complementary differential (pair of npn for +voltage and pair of pnp for -voltage) makes more sense if we see them. It makes the whole circuit more reasonable, and more symmetrical. Upper differential is working for all upper transistors, and lower differential is working for the negative half. Especially if we look at high power amplifier (pro-audio) where the rail can reach more than +/-100VDC, maybe this complementary differential is the only answer, to divide the working voltage (Vce) only from +rail to ground, not to -rail. This is due to the dissipation limitation with certain current we must handle (and also limitation from maximum Vce for transistors)
But when I notice the comment from Mr.Douglas Self, and see the design of Mr.Nelson Pass, this question rises in my mind. Mr Pass once told me to read the article he wrote about how mosfets works. In his paper, I learned that N mosfet and P mosfets have very different Vgs thereshold. And when I read the history of bipolar transistors, I also learned that PNP and NPN are very different. NPN is what we found first. It is the nature of transistors. PNP is found long time after NPN, and using very different moulding from NPN. We can see from datasheets, especially mosfets (like IRF540 and IRF9540), the N mosfets always superior in datasheet than P mosfets (with the same shape).
In power amp, differential stage is what I think the most important stage in shaping the output. It works with small signal, and everything began here. So every little difference or mismatch will be sighted as all the signal amplifies. We can cover some of this mismatch by putting feedback, forcing to fix the errors. But the question is "which is better?"
So if the NPN and PNP are not the exact oposite (with exact behavior) will my original question here makes sense?
But when I notice the comment from Mr.Douglas Self, and see the design of Mr.Nelson Pass, this question rises in my mind. Mr Pass once told me to read the article he wrote about how mosfets works. In his paper, I learned that N mosfet and P mosfets have very different Vgs thereshold. And when I read the history of bipolar transistors, I also learned that PNP and NPN are very different. NPN is what we found first. It is the nature of transistors. PNP is found long time after NPN, and using very different moulding from NPN. We can see from datasheets, especially mosfets (like IRF540 and IRF9540), the N mosfets always superior in datasheet than P mosfets (with the same shape).
In power amp, differential stage is what I think the most important stage in shaping the output. It works with small signal, and everything began here. So every little difference or mismatch will be sighted as all the signal amplifies. We can cover some of this mismatch by putting feedback, forcing to fix the errors. But the question is "which is better?"
So if the NPN and PNP are not the exact oposite (with exact behavior) will my original question here makes sense?
> high power amplifier (pro-audio) where the rail can reach more than +/-100VDC, maybe this complementary differential is the only answer, to divide the working voltage (Vce) only from +rail to ground, not to -rail.
Nope. Same voltage either way.
> the history of bipolar transistors, I also learned that PNP and NPN are very different. NPN is what we found first. It is the nature of transistors. PNP is found long time after NPN
Sounds like a highly simplified history. From 1954 to 1964, THE most common transistor was the Germanium, and 99% of those were PNP, from computers to telephone switchgear to car and pocket radios. The reason "all" cars (not just Cadillac) went to 12V Negative Ground was so we could bolt the collector of a monster Ge PNP to the radio case for cooling. Then came Silicon, which is slightly easier to make in NPN. But both polarities of both materials were available from the beginning, and there is an exhaustive survey of complementary topologies in a book I have that was published just a few years after the transistor was invented.
But history does not matter. Yes, a PNP BPJ has slightly higher ohmic loss and Vbe than an NPN BJT of the same size, but the difference is very small and you can find pairs of devices where the difference from NPN to PNP is less than the difference from one transistor to the next of the same type. In MOSFETs the difference is larger, enough so you "never" use a P-type for switching if you can avaoid it. But in linear or AB amplifiers, you always add some fixed resistance (or other feedback) to set bias current and reduce nonlinearity. This has to be larger than the devices' own internal resistance, so the difference between the P-side and the N-side becomes very small.
> differential stage is what I think the most important stage in shaping the output. It works with small signal
There are many ways to look at amplifiers, and you do have to look at the design from all directions.
But in one big way, the input stage is the "easiest". The output devices have to swing current from 0 Amps to 5 Amps, and large voltage swing too. The input devices can be designed (by having plenty of current gain in the whole amp) to swing only from 99µA to 101µA with nearly constant voltage (and low voltage if you want). While they have a very critical job, they do not have to carry the strain of pushing big current or even large current changes. In principle you can make the signal across the input pair(s) as small as you like, and push distortion down to the vanishing point and accuracy up as high as you could ever hope for.
Nope. Same voltage either way.
> the history of bipolar transistors, I also learned that PNP and NPN are very different. NPN is what we found first. It is the nature of transistors. PNP is found long time after NPN
Sounds like a highly simplified history. From 1954 to 1964, THE most common transistor was the Germanium, and 99% of those were PNP, from computers to telephone switchgear to car and pocket radios. The reason "all" cars (not just Cadillac) went to 12V Negative Ground was so we could bolt the collector of a monster Ge PNP to the radio case for cooling. Then came Silicon, which is slightly easier to make in NPN. But both polarities of both materials were available from the beginning, and there is an exhaustive survey of complementary topologies in a book I have that was published just a few years after the transistor was invented.
But history does not matter. Yes, a PNP BPJ has slightly higher ohmic loss and Vbe than an NPN BJT of the same size, but the difference is very small and you can find pairs of devices where the difference from NPN to PNP is less than the difference from one transistor to the next of the same type. In MOSFETs the difference is larger, enough so you "never" use a P-type for switching if you can avaoid it. But in linear or AB amplifiers, you always add some fixed resistance (or other feedback) to set bias current and reduce nonlinearity. This has to be larger than the devices' own internal resistance, so the difference between the P-side and the N-side becomes very small.
> differential stage is what I think the most important stage in shaping the output. It works with small signal
There are many ways to look at amplifiers, and you do have to look at the design from all directions.
But in one big way, the input stage is the "easiest". The output devices have to swing current from 0 Amps to 5 Amps, and large voltage swing too. The input devices can be designed (by having plenty of current gain in the whole amp) to swing only from 99µA to 101µA with nearly constant voltage (and low voltage if you want). While they have a very critical job, they do not have to carry the strain of pushing big current or even large current changes. In principle you can make the signal across the input pair(s) as small as you like, and push distortion down to the vanishing point and accuracy up as high as you could ever hope for.
You folks are worrying too much. The complementary differential has lower distortion, all else being equal. If you build a design like the comp symmetry example shown, just put a large cap across the second stage base to the nearest supply. Either side will do. I found an increase of distortion of 5 times in an example that I measured more than 30 years ago.
Think it through. First you lose GAIN (6db) because you have only one working input stage. Second, you INCREASE even harmonic production, because you are not equally driving the second stage transistors. Try it and see. Don't worry about the intrinsic mismatch in N vs P transistors or fets, it still is better to use both together.
Think it through. First you lose GAIN (6db) because you have only one working input stage. Second, you INCREASE even harmonic production, because you are not equally driving the second stage transistors. Try it and see. Don't worry about the intrinsic mismatch in N vs P transistors or fets, it still is better to use both together.
> You folks are worrying too much.
Who is worrying? lumanauw has an idea that one topology is better than another. In the hands of a skilled experienced designer, this may be true (though all-else is never equal). But at lumanauw's apparent level of learning, it is important to realize that no topology is "unavoidably" better than another. We can easily design a "dual differential" that is bad (heck, I've done that), and a "single differential" that is better. Many sing-diff inputs work very well. Complementary diff has some "obvious" advantages, but they can be botched with careless detailing. lumanauw's theory may be right. But he needs to understand that the devil is in the details.
That's where Doug Self is useful. He has worried the Classic Topology to the bone, and pointed out many "small" details where most designs could be significantly better. The concept and many of the details can be applied to other topologies. But you have to learn them before you can apply them. Self teaches this well. He gets fabulous performance out of something that looks like a 1972 Fisher. We can criticize that he works mostly by measurements, not by ear. But his details look like things that "should" be better to the ear, and others with golden-ears can judge his work for themselves. There are details where I disagree with his theory, but he is honestly investigating and sharing which is all anybody can do.
> you lose GAIN (6db)
Too true. Though gain in BJT is cheap. And with overall feedback and slow output devices, gain has to droop inside the audio band to keep the loop stable. That -tends- to lead to a design where almost all gain is made in one stage, usually the second stage, so the total gain can be compensated to one-pole response. With the huge potential gain of BJTs, this usually means deliberate reduction of input stage gain with emitter resistors.
FETs is different. Gain is harder to come by. In a sense, they already have the "emitter resistors" built-in that we add to many BJT stages. If you can find semi-matched complements (I sure agree that small mismatching is no big deal) then all-complementary is one preferred topology.
> you INCREASE even harmonic production, because you are not equally driving the second stage transistors.
Less if you detail so the standing current is very-large compared to the signal current (base current in the output stage). This is one place where many BJT designs come up a dollar short.
I'm also, at my current level of understanding, becoming less concerned about THD and more about distortion spectrum. A spectrum with no even-order is un-natural. If the slope of the spectrum is steep and/or all products are far below system noise, that won't matter. And I'm not sure I can equate lumanauw's observation "more focus" with distortion spectrum.
Who is worrying? lumanauw has an idea that one topology is better than another. In the hands of a skilled experienced designer, this may be true (though all-else is never equal). But at lumanauw's apparent level of learning, it is important to realize that no topology is "unavoidably" better than another. We can easily design a "dual differential" that is bad (heck, I've done that), and a "single differential" that is better. Many sing-diff inputs work very well. Complementary diff has some "obvious" advantages, but they can be botched with careless detailing. lumanauw's theory may be right. But he needs to understand that the devil is in the details.
That's where Doug Self is useful. He has worried the Classic Topology to the bone, and pointed out many "small" details where most designs could be significantly better. The concept and many of the details can be applied to other topologies. But you have to learn them before you can apply them. Self teaches this well. He gets fabulous performance out of something that looks like a 1972 Fisher. We can criticize that he works mostly by measurements, not by ear. But his details look like things that "should" be better to the ear, and others with golden-ears can judge his work for themselves. There are details where I disagree with his theory, but he is honestly investigating and sharing which is all anybody can do.
> you lose GAIN (6db)
Too true. Though gain in BJT is cheap. And with overall feedback and slow output devices, gain has to droop inside the audio band to keep the loop stable. That -tends- to lead to a design where almost all gain is made in one stage, usually the second stage, so the total gain can be compensated to one-pole response. With the huge potential gain of BJTs, this usually means deliberate reduction of input stage gain with emitter resistors.
FETs is different. Gain is harder to come by. In a sense, they already have the "emitter resistors" built-in that we add to many BJT stages. If you can find semi-matched complements (I sure agree that small mismatching is no big deal) then all-complementary is one preferred topology.
> you INCREASE even harmonic production, because you are not equally driving the second stage transistors.
Less if you detail so the standing current is very-large compared to the signal current (base current in the output stage). This is one place where many BJT designs come up a dollar short.
I'm also, at my current level of understanding, becoming less concerned about THD and more about distortion spectrum. A spectrum with no even-order is un-natural. If the slope of the spectrum is steep and/or all products are far below system noise, that won't matter. And I'm not sure I can equate lumanauw's observation "more focus" with distortion spectrum.
I would like to make the case for complementary differential topology:
It isn't the INPUT stage that is lowered in distortion, it is the 2'nd stage which is usually single ended and has to develop almost all the gain for the amp, which improves. This problem was first addressed with 'bootstrapping' using a cap connected to the output of the amp to give positive feedback and increase the driver load impedance. The next approach was to use a constant current source as a load, favored even today by Doug Self. Finally, the equal driving of both driver transistors, either with a current mirror, or with a complementary differential input.
I have used each of these approaches over the last 35 years, and personally, I prefer the complementary differential fet input. Don't tell Doug Self, but fets actually work darn well as input stages, and have many advantages, such as no need for an input capacitor, and very high slew rate operation, without any noise tradeoff. Also, they tend to be more RFI resistant, because their input diode is off, rather than conducting.
While I have the greatest respect for what Doug Self has published, please don't box yourself in a corner by thinking that that his input is the only or necessarily the best approach to circuit design.
It isn't the INPUT stage that is lowered in distortion, it is the 2'nd stage which is usually single ended and has to develop almost all the gain for the amp, which improves. This problem was first addressed with 'bootstrapping' using a cap connected to the output of the amp to give positive feedback and increase the driver load impedance. The next approach was to use a constant current source as a load, favored even today by Doug Self. Finally, the equal driving of both driver transistors, either with a current mirror, or with a complementary differential input.
I have used each of these approaches over the last 35 years, and personally, I prefer the complementary differential fet input. Don't tell Doug Self, but fets actually work darn well as input stages, and have many advantages, such as no need for an input capacitor, and very high slew rate operation, without any noise tradeoff. Also, they tend to be more RFI resistant, because their input diode is off, rather than conducting.
While I have the greatest respect for what Doug Self has published, please don't box yourself in a corner by thinking that that his input is the only or necessarily the best approach to circuit design.
This is an extremely well-informed thread. Comments from Jonathan, PRR and John are outstanding.
I agree strongly with PRR's comment about distortion spectrum and its affect on musical perception. Distortion spectrum is, in my opinion, the crucial factor; you want something with maximal H2, progressively decreasing with increasing harmonics, and least, if not zero, from H6 onwards. The total distortion figure is not too important, though intermodulation performance is VERY important. The significant thing is to have a distortion spectrum which is not too dissimilar from that found in the natural world. High levels of odd order, with evens missing, or very small, is not natural sounding.
A complementary dual diff will greatly minimize H2, H4, H6 - even order - while not much increasing H3, H5, H7 - odd order. A single diff will have quite high H2, some H3, some H4, and some H5. This observation is counter-intuitive because a complementary dual diff 'looks' right. The point concerning driving a fully symmetrical voltage amplifier (more correctly described as a transresistance amp as it converts current input to voltage output) is well taken. There are linearity advantages to a fully complementary VAS, but, as before, even order harmonic generation is nulled.
I agree with John that pretty much any topology can be made to sound good. The schematic is just the beginning; it takes careful dimensioning, component choice and layout to produce good sonics, but it is possible voice almost anything to sound good. Good sonics are probably equal part topology, dimensioning, component choice and layout, and a huge amount of work is required to make it happen. You will NEVER know by simply examining the schematic, any more than you can definitively gauge the emotional impact of a symphony by studying the score. My personal take is for a progressively reducing harmonic spectrum, adjusted so H2 predominates, but with very careful 'voicing' which is really only achievable with lots of listening. It is also true that some people like the SE sound, while others the PP sound. This is even true of Class A versus AB; for myself, I prefer Class AB as it seems somehow more dynamic and lifelike, at least in my designs. You can't please 'em all, and you'll die trying!
We tend to concentrate on the technology in straight electrical terms, but much work remains to be done on the psycho-acoustic phenomenon, the subjective aspects of why we like what we like, and why we tend to become religious zealots about it. More work is also required in the area of harmonic spectrum, particularly as it relates to musical scales, and profiling musical instrument tones. Much of this work is actually known, but for some reason it does not seem to be found in the audiophile and designer communities.
Cheers,
Hugh
I agree strongly with PRR's comment about distortion spectrum and its affect on musical perception. Distortion spectrum is, in my opinion, the crucial factor; you want something with maximal H2, progressively decreasing with increasing harmonics, and least, if not zero, from H6 onwards. The total distortion figure is not too important, though intermodulation performance is VERY important. The significant thing is to have a distortion spectrum which is not too dissimilar from that found in the natural world. High levels of odd order, with evens missing, or very small, is not natural sounding.
A complementary dual diff will greatly minimize H2, H4, H6 - even order - while not much increasing H3, H5, H7 - odd order. A single diff will have quite high H2, some H3, some H4, and some H5. This observation is counter-intuitive because a complementary dual diff 'looks' right. The point concerning driving a fully symmetrical voltage amplifier (more correctly described as a transresistance amp as it converts current input to voltage output) is well taken. There are linearity advantages to a fully complementary VAS, but, as before, even order harmonic generation is nulled.
I agree with John that pretty much any topology can be made to sound good. The schematic is just the beginning; it takes careful dimensioning, component choice and layout to produce good sonics, but it is possible voice almost anything to sound good. Good sonics are probably equal part topology, dimensioning, component choice and layout, and a huge amount of work is required to make it happen. You will NEVER know by simply examining the schematic, any more than you can definitively gauge the emotional impact of a symphony by studying the score. My personal take is for a progressively reducing harmonic spectrum, adjusted so H2 predominates, but with very careful 'voicing' which is really only achievable with lots of listening. It is also true that some people like the SE sound, while others the PP sound. This is even true of Class A versus AB; for myself, I prefer Class AB as it seems somehow more dynamic and lifelike, at least in my designs. You can't please 'em all, and you'll die trying!
We tend to concentrate on the technology in straight electrical terms, but much work remains to be done on the psycho-acoustic phenomenon, the subjective aspects of why we like what we like, and why we tend to become religious zealots about it. More work is also required in the area of harmonic spectrum, particularly as it relates to musical scales, and profiling musical instrument tones. Much of this work is actually known, but for some reason it does not seem to be found in the audiophile and designer communities.
Cheers,
Hugh
Mr.John Curl, I agree that anyone cannot take only 1 resource as the only right thing, especially in audio electronics. With more discussion and learning from various experienced people we can get closer to what is the right fact. Actually I always wanted to try to built fet differential, like using k389 or j109. But they are difficult to get here. The best I can get is k30, but that is only 30vmax. Do you have any example of your preferred fet differential input for me to see?
Mr.PRR. I'm interested in using PC's computer card as Harmonic distortion analyzer. In another thread you have mentioned it, but for me it is not clear enough. Can you give me any url on how can I built myself a distortion analyzer with PC's card? Is there any additional software needed?
I'm getting confused when people talks about 2nd order, 3rd, 4th, 5th order harmonics, etc.
Are they (every single harmonics) visible at the pc if we use your arrangement of PC distortion analyzer? Or is it just numbers from calculation?
Mr.PRR. I'm interested in using PC's computer card as Harmonic distortion analyzer. In another thread you have mentioned it, but for me it is not clear enough. Can you give me any url on how can I built myself a distortion analyzer with PC's card? Is there any additional software needed?
I'm getting confused when people talks about 2nd order, 3rd, 4th, 5th order harmonics, etc.
Are they (every single harmonics) visible at the pc if we use your arrangement of PC distortion analyzer? Or is it just numbers from calculation?
john curl said:
What it doesn't mean is that we're more sensitive to odd-ordered harmonics, which is what Hugh had claimed. Please read what he wrote:
This argument is largely based on the fact that the ear is not sensitive to large levels of H2, up to about 2%, yet hypersensitive to odd, higher orders, such as H7 (around 0.05% according to studies I have seen of the differences between soft and hard trumpet sounds.)
He's not saying that odd-order harmonics sound different compared to even-order harmonics. He's saying that we're more sensitive to odd-order harmonics compared to even-order harmonics.
Well I'm ignoring it here because it hasn't anything to do with the issue in question. And the issue isn't whether odd-order harmonics sound different from even-order harmonics, but rather whether odd-order harmonics are more audible compared to even-order harmonics.
SE has told me, in print, to sell my books, since my personally having them is a waste of time and space. Here is one exception to his assertion.
Yes, I'd said that they're a waste of space because of your habit of supporting your arguments by telling people how many books you have on the shelf (i.e. "You're wrong! I've got 200 books on my shelf so I know what I'm talking about!") instead of actually quoting something from those books to support your argument.
So here I'll give you credit for actually quoting something from those books, but unfortunately what you quoted hasn't anything to do with the question at hand.
se
Re: Trying to get back on track...
Over the years many designers have made different choices of details in the differential input stage alone; Fet vs Bjt, degeneration, cascodes (Fet, Bjt, bootstrapped, folded), current source vs R bias, current mirror vs resistive load, device part #, manufacturers, matching – any of which could be considered a “detail” of implementation that doesn’t rise to the level of the question of whether to use them in a single or complementary differential input
for a "light" (if not ultimately enlightening) discussion:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=17524&perpage=15&highlight=&pagenumber=1
I'm afraid that this question has too many dimensions to be answered in any useful way, I would emphasize PRR's and others comments that implementation "details" totally trump any generalizations drawn from simplistic heuristics and visually pleasing symmetries in the schematic representationSawzall said:
Over the years many designers have made different choices of details in the differential input stage alone; Fet vs Bjt, degeneration, cascodes (Fet, Bjt, bootstrapped, folded), current source vs R bias, current mirror vs resistive load, device part #, manufacturers, matching – any of which could be considered a “detail” of implementation that doesn’t rise to the level of the question of whether to use them in a single or complementary differential input
for a "light" (if not ultimately enlightening) discussion:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=17524&perpage=15&highlight=&pagenumber=1
> harmonics created by the machine perform minimal spectral shift on the sounds which are being processed.
Not that simple.
The real harmonics are in the ear. Huge levels of 2nd harmonic. It is irrelevant how much 2nd an amplifier makes. Even on clarinet. And as Steve points out, musical instrument harmonics are not perfect multiples of the fundamental or each other (piano is worst, but even a clarinet has inharmonic overtones).
I don't think that cancelling the even-order harmonics is good or bad in itself. Every-harmonic, or every-other-harmonic, doesn't seem to make a lot of difference. Curl suggests that he used to null even-order, probably because "he could", and has shifted away from that, perhaps because it makes little difference to his designs and his ears, or he finds other benefits in not-nulling the evens.
What does seem to matter is getting the spectrum to slope steeply, and/or get it well below noise level.
No-feedback triodes have a steep slope of distortion spectrum. It roughly parallels the ear's own distortion and tends to be masked.
High-feedback amps (any technology and topology that allows really high feedback) can get distortion well below system noise where it vanishes. (It has to be well below system noise; we hear tones 10dB or so below broadband noise.) That works well enough in "small" amps which can be run well below clipping. In large power amps, transient overload non-ideal behavior may color the sound even when "perfect on test tones".
> 40dB per octave
That's a steeper slope than I think is needed. I'm not sure the masking threshold is the right guide. But for comparision: simple THD measurements apply NO weighting and are well-known to be inadequate. IMD applies a first-order weighting and often compares better to "sound". Olsen's tests suggest more like a second-order weighting (12dB/8ve, 40dB/decade), and RDF 4th cites a study where this is shown to corellate well, but this seems to be forgotten and certainly needs re-examination on current "very clever" amplifers.
As to whether hollow-state, BJT, FET, etc is "best": aside from some real-life issues (you can't take huge feedback around a tube with the transformer it often wants), I think very-good designs are possible many ways. John argues well for FETs; a hard-core BJT designer could argue against his points. I'm inclined to think that details are more important than devices. And also: that many existing designs have flawed details. Case in point: the many straight and elaborated "Classic BJT power amplifiers" that don't work as well as Doug Self's designs. There may be better amps than Self's, but there are certainly a lot of worse ones out there.
Bootstrapping an AB output.... shudder. I think it was whatzhisname at SWTP who first woke me up to this folly. I mentioned the 1972 Fisher because it was a clean example of the post-bootstrap era.
> what effects would the various topologies have on the error spectrum?
Less effect than the detailing. If a BJT diff-pair is not kept perfectly balanced, distortion won't cancel. Not keeping enough effective current gain does nasty things to BJTs. And while Class AB output stages are a mess, there are a lot of bad Class A designs too. Some of these ills are "easier to avoid" with FETs, but you can't just stick some FETs in a "topology" and get great results.
As for "getting stuck in Doug Self"... While I only discovered Self a few years ago, I've been mucking-around with The Classic Topology for (hmmm...) 30 years, and still find tidbits to chew on in his writings. More brilliant designers have noted all this before and moved on, but for many of us a few years of pondering Doug Self is a good foundation for design in any device or topology.
> 'Science and Music' Sir James Jeans 1937.... "... The seventh harmonic, however,..."
Same thing pointed out by Helmholtz in On the Sensations of Tone written in 1875. Before there were amplifiers and harmonic meters. It is amazing how much he demonstrated, measured and extrapolated without electricity. It is heavy reading, and some of the terminology has changed since then, but if you really are interested in what and how we hear you must start reading it.
Another classic, mostly for musicians, is Benade's Fundamentals of Musical Acoustics.
All these books are shockingly cheap via Dover's reprints. (I don't want to know how much my near-mint 1882 copy of Helmholtz is worth...)
> some folks should just be ignored
At least not over-reacted-to.
Not that simple.
The real harmonics are in the ear. Huge levels of 2nd harmonic. It is irrelevant how much 2nd an amplifier makes. Even on clarinet. And as Steve points out, musical instrument harmonics are not perfect multiples of the fundamental or each other (piano is worst, but even a clarinet has inharmonic overtones).
I don't think that cancelling the even-order harmonics is good or bad in itself. Every-harmonic, or every-other-harmonic, doesn't seem to make a lot of difference. Curl suggests that he used to null even-order, probably because "he could", and has shifted away from that, perhaps because it makes little difference to his designs and his ears, or he finds other benefits in not-nulling the evens.
What does seem to matter is getting the spectrum to slope steeply, and/or get it well below noise level.
No-feedback triodes have a steep slope of distortion spectrum. It roughly parallels the ear's own distortion and tends to be masked.
High-feedback amps (any technology and topology that allows really high feedback) can get distortion well below system noise where it vanishes. (It has to be well below system noise; we hear tones 10dB or so below broadband noise.) That works well enough in "small" amps which can be run well below clipping. In large power amps, transient overload non-ideal behavior may color the sound even when "perfect on test tones".
> 40dB per octave
That's a steeper slope than I think is needed. I'm not sure the masking threshold is the right guide. But for comparision: simple THD measurements apply NO weighting and are well-known to be inadequate. IMD applies a first-order weighting and often compares better to "sound". Olsen's tests suggest more like a second-order weighting (12dB/8ve, 40dB/decade), and RDF 4th cites a study where this is shown to corellate well, but this seems to be forgotten and certainly needs re-examination on current "very clever" amplifers.
As to whether hollow-state, BJT, FET, etc is "best": aside from some real-life issues (you can't take huge feedback around a tube with the transformer it often wants), I think very-good designs are possible many ways. John argues well for FETs; a hard-core BJT designer could argue against his points. I'm inclined to think that details are more important than devices. And also: that many existing designs have flawed details. Case in point: the many straight and elaborated "Classic BJT power amplifiers" that don't work as well as Doug Self's designs. There may be better amps than Self's, but there are certainly a lot of worse ones out there.
Bootstrapping an AB output.... shudder. I think it was whatzhisname at SWTP who first woke me up to this folly. I mentioned the 1972 Fisher because it was a clean example of the post-bootstrap era.
> what effects would the various topologies have on the error spectrum?
Less effect than the detailing. If a BJT diff-pair is not kept perfectly balanced, distortion won't cancel. Not keeping enough effective current gain does nasty things to BJTs. And while Class AB output stages are a mess, there are a lot of bad Class A designs too. Some of these ills are "easier to avoid" with FETs, but you can't just stick some FETs in a "topology" and get great results.
As for "getting stuck in Doug Self"... While I only discovered Self a few years ago, I've been mucking-around with The Classic Topology for (hmmm...) 30 years, and still find tidbits to chew on in his writings. More brilliant designers have noted all this before and moved on, but for many of us a few years of pondering Doug Self is a good foundation for design in any device or topology.
> 'Science and Music' Sir James Jeans 1937.... "... The seventh harmonic, however,..."
Same thing pointed out by Helmholtz in On the Sensations of Tone written in 1875. Before there were amplifiers and harmonic meters. It is amazing how much he demonstrated, measured and extrapolated without electricity. It is heavy reading, and some of the terminology has changed since then, but if you really are interested in what and how we hear you must start reading it.
Another classic, mostly for musicians, is Benade's Fundamentals of Musical Acoustics.
All these books are shockingly cheap via Dover's reprints. (I don't want to know how much my near-mint 1882 copy of Helmholtz is worth...)
> some folks should just be ignored
At least not over-reacted-to.
Who was that masked man?
"Less effect than the detailing"
What an excellent post. I may nominate it for post of the week. There so many considerations in designing a good amp that the choice of the front end topology is a small part of the equation. I have heard several really nice sounding amps that had very different front end topologies. I am building an amp (a friends design) with a very simple and classic topology but with great attention to details. I am very curious what an "oldy but goody" design sounds like with attention to details.....
"Less effect than the detailing"
What an excellent post. I may nominate it for post of the week. There so many considerations in designing a good amp that the choice of the front end topology is a small part of the equation. I have heard several really nice sounding amps that had very different front end topologies. I am building an amp (a friends design) with a very simple and classic topology but with great attention to details. I am very curious what an "oldy but goody" design sounds like with attention to details.....
PRR said:
Yes, we can hear pure tones a good way down below broadband noise. But the music itself is rather broadband and will be dramatically greater than the system noise so the question becomes how deep into the music can we hear under practical conditions?
Under practical conditions we have not only system noise, but the "noise" of the music itself, as well as the ambient noise in the listening room which even in a recording studio is typically around 30dB SPL, not to mention the fact that the typical listening room is anything but anechoic so you also have a rather reverberant acoustic environent to contend with.
And then there are our ears themselves which when exposed to sound have an autonomic "clinching" response to sound such that the very act of listening to music at typical listening levels dramatically reduces their sensitivity which is already reduced due to the ambient noise levels we're constantly exposed to.
For example, entering an anechoic chamber from even a relatively quiet ambient environment, it takes some time before our ears "relax" and fully acclimate to the reduced noise level before you can start hearing all those things you don't otherwise hear.
An excerpt from Everest's Master Handbook of Acoustics illustrates this well:
The delicate and sensitive nature of our hearing can be underscored dramatically by a little experiment. A bulky door of an anechoic chamber is slowly opened, revealing extremely thick walls, and three-foot wedges of glass fiber, points inward, lining all walls, ceiling, and what coul dbe called a floor, except that you walk on an open steel grillwork.
A chair is brought in, and you sit down. This experiment takes time, and as a result of prior briefing, you lean back, patiently counting the glass fiber wedges to pass the time. It is very eerie in here. The sea of sound and noises of life and activity in which we are normally immersed and of which we are ordinarily scarcely conscious is now conspicuous by its absence.
The silence presses down on you in the tomblike silence, 10 minutes, then a half hour pass. New sounds are discovered, sounds that come from within your own body. First, the loud pounding of your heart, still recovering from the novelty of the situation. An hour goes by. The blood coursing through the vessels becomes audible. At last, if your ears are keen, your patience is rewarded by a strange hissing sound between the "ker-bumps" of the heart and the slushing of blood. What is it? It is the sound of air particles pounding against your eardrums.
So now how sensitive are our ears after being exposed for a time to music at average levels on the order of 80-90dB SPL, on top of the ambient noise, and in a reverberant environment?
In any case, I don't believe there is any singular "best" way to go about designing an amplifier. At least in the practical sense, outside of purely technical arguments. Most every approach has its adherents and the subjective tastes and preference of listeners varies considerably which is why everything from highly non-linear single-ended tube amps and highly linear solid state amps such as the Halcros are able to thrive in the same market.
I think it ultimately just boils down to trying things for yourself and decide which you ultimately prefer.
se
> I'm interested in using PC's computer card as Harmonic distortion analyzer. Can you give me any url on how can I built myself a distortion analyzer with PC's card?
RMAA 5.1 release with manual on "Basics of Audio Measurements"
> 2nd order, 3rd, 4th, 5th order harmonics, etc. Are they (every single harmonics) visible at the pc...?
Yes, and audible to the ear. Get a sine wave generator. Put it through an amplifier. At normal levels the output sounds a lot like the input. When you crank it "too high", it sounds like another higher-pitched instrument has joined-in. If that instrument is playing an octave higher (2X frequency) than the fundamental, it is Second Harmonic. If it sounds like 1.5 octaves higher (3X frequency), it is Third Harmonic. Two octaves (4X freq) is Fourth Harmonic.
(Note that Physicists count the Fundamental as "First", while musicians often speak of "partials" counting the octave-up overtone as "first". Most technical literature uses the Physicists' numbering, but you may see it otherwise in musician's and instrument-makers' writings.)
In a single-ended tube amplifier, you typically hear the 2nd come up first, then the 3rd, and if you listen close you can identify 4th, 5th, and higher harmonics. Some push-pull or heavy-feedback amplifiers won't make the even-numbered harmonics no matter how hard you smack them. BUt many-many "push pull" amplifiers have non-push-pull driver stages. Curl cites this as a reason to go all push-pull. If you use the simple push-pull tube amplifier with voltage-amp, split-load phase inverter, and P-P output, with triodes, the driver stage is often making lots of 2nd harmonic while the output's 3rd harmonic is still small.
While it is easiest to hear these harmonics when you smack an amp into clipping, lesser amounts of harmonics are present all the time, and you hear them even if you can't pick them out. Once you hear them "big", it is easier to identify small amounts of distortion.
RMAA 5.1 release with manual on "Basics of Audio Measurements"
> 2nd order, 3rd, 4th, 5th order harmonics, etc. Are they (every single harmonics) visible at the pc...?
Yes, and audible to the ear. Get a sine wave generator. Put it through an amplifier. At normal levels the output sounds a lot like the input. When you crank it "too high", it sounds like another higher-pitched instrument has joined-in. If that instrument is playing an octave higher (2X frequency) than the fundamental, it is Second Harmonic. If it sounds like 1.5 octaves higher (3X frequency), it is Third Harmonic. Two octaves (4X freq) is Fourth Harmonic.
(Note that Physicists count the Fundamental as "First", while musicians often speak of "partials" counting the octave-up overtone as "first". Most technical literature uses the Physicists' numbering, but you may see it otherwise in musician's and instrument-makers' writings.)
In a single-ended tube amplifier, you typically hear the 2nd come up first, then the 3rd, and if you listen close you can identify 4th, 5th, and higher harmonics. Some push-pull or heavy-feedback amplifiers won't make the even-numbered harmonics no matter how hard you smack them. BUt many-many "push pull" amplifiers have non-push-pull driver stages. Curl cites this as a reason to go all push-pull. If you use the simple push-pull tube amplifier with voltage-amp, split-load phase inverter, and P-P output, with triodes, the driver stage is often making lots of 2nd harmonic while the output's 3rd harmonic is still small.
While it is easiest to hear these harmonics when you smack an amp into clipping, lesser amounts of harmonics are present all the time, and you hear them even if you can't pick them out. Once you hear them "big", it is easier to identify small amounts of distortion.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.