It means eliminating (not just reducing, altering, or replacing it with another non-linearity) crossover distortion in an amp that's operating outside the Class A bias region and delivering significantly better efficiency than you can get with a pure Class A design. Or put another way, an amp that measures as good (or better) than a pure Class A design but is more efficient.
Most "magic bias" approaches I've seen either start with Class A and try to improve the efficiency with various methods. Or they start with Class B and try to reduce (or some claim to eliminate) the crossover distortion. Either approach is usually only partially effective and typically has some significant downside. So with any practical approaches I've seen, you are in effect, trading one problem for another.
For some a magic bias design might be a worthwhile trade off for various reasons. But, objectively, I've not seen any schemes that can match the distortion performance of a simple pure Class A design.
Most "magic bias" approaches I've seen either start with Class A and try to improve the efficiency with various methods. Or they start with Class B and try to reduce (or some claim to eliminate) the crossover distortion. Either approach is usually only partially effective and typically has some significant downside. So with any practical approaches I've seen, you are in effect, trading one problem for another.
For some a magic bias design might be a worthwhile trade off for various reasons. But, objectively, I've not seen any schemes that can match the distortion performance of a simple pure Class A design.
You don't even need a magic bias to beat most class A amps. This link is a good example of that. Have you ever seen a class A amp with specs better than this?
http://www.diyaudio.com/forums/showthread.php?s=&threadid=147759
Steve Dunlap said:
I agree the "YAP" appears to be a great amp. It uses one of my favorite op-amps and among my favorite output devices. I do question, however, that it can really manage 0.0001% THD at 200 watts and 20 Khz (no offense to syn08). That's an unheard of number in any published data I've ever seen. But even if his amp is an order of magnitude worse, it would still be a great amp. I'd love to see a full set of AP data on it.
And I agree one can argue the end result matters more than the means (or bias class in this case). I'm not trying to say minute differences in THD are what matter most. But, when it comes to evaluating bias classes and output stage designs--the topic of this thread--I don't know how else to objectively quantify one design over another.
When evaluating residual distortion, it's also worth noting that crossover distortion is often invisible unless you average the residual over many samples to "extract" the repetitive distortion component from the random stuff. And how audible it really is when playing music depends on who you ask.
I like to use differencing to evaluate power amps (i.e. per Baxendall, Hafler, Waslo, et al) playing real music into typical real speakers. It's my "acid test" and works very well for comparing changes to a given design or comparing two different amps side by side under the same conditions. Unfortunately the technique, unlike say AP measurements, does *not* lend itself well to comparing results obtained with different test set ups. So I can't accurately compare my differencing data with Waslo's for example.
But if you set out, like Bigun, to design an amp you can confidently listen to (up to some level anyway) knowing it's free of crossover distortion, Class A seems like the obvious choice. Syn08's YAP design could deliver better performance as there obviously are many variables involved. The only way I know to find out is to test real hardware in a consistent way on a level playing field.
RocketScientist said:
The spec is probably from a simulation.
Simulations are notorious for being in accurate.
Its how the amp performs in the real world that is most important.
Personally I wouldnt touch a class a becasue I run a mobile disco and play guitar in pubs, this means I need many hundreds of watts. I am perfectly happy with my own design MOSFET class AB amplifier.
RS,
For a man convinced that rigorous measurement is the only way to really differentiate amps, I'm very surprised you put any faith at all in listening tests.
Are you now suggesting, after regaling us with 'one man's pleasure is another mans' poison', that the listening test is the only way to go?
I enjoyed your analogy about choosing a car from an artist's impression. I think that's closer to the mark. But since distortion - and hence amp choice - is such a subjective thing, would you not say that choosing an amp is a highly individual thing, and should be done with listening tests only (despite favorite choices of output devices) by the person wishing to buy for best effect?
I notice you have not even referred to distortion spectrums, the work done by Earle Geddes, Lee, and even Jean Hiraga. Surely this rates a mention?
If you remove crossover distortion, you eliminate one more distortive problem in amps. This is laudable, not contemptible, and contributes to the art. If you replace it with another distortion (though I can't see how this might happen) and if it occurs at crossover, surely it too must be crossover distortion, and the original intent is lost?
Have you ever built and compared, both measured and subjective, Class AB amps of varying bias depth and Class A? A few examples might serve your arguments/opinions well. The art is furthered only by correlating topology, measured distortion, and subjective experience, and opinions on the measured performance alone regrettably do not complete the picture. It appears to be much more complex than thought.
BTW, some highly distortive tube circuits subjectively score very highly, even amongst credentialled engineers. Other very low distortion SS amps sound terrible. NP's single ended Class A circuits sound wonderful. Where is the correlation with the measured distortion? Can you explain this to us?
Cheers,
Hugh
For a man convinced that rigorous measurement is the only way to really differentiate amps, I'm very surprised you put any faith at all in listening tests.
Are you now suggesting, after regaling us with 'one man's pleasure is another mans' poison', that the listening test is the only way to go?
I enjoyed your analogy about choosing a car from an artist's impression. I think that's closer to the mark. But since distortion - and hence amp choice - is such a subjective thing, would you not say that choosing an amp is a highly individual thing, and should be done with listening tests only (despite favorite choices of output devices) by the person wishing to buy for best effect?
I notice you have not even referred to distortion spectrums, the work done by Earle Geddes, Lee, and even Jean Hiraga. Surely this rates a mention?
If you remove crossover distortion, you eliminate one more distortive problem in amps. This is laudable, not contemptible, and contributes to the art. If you replace it with another distortion (though I can't see how this might happen) and if it occurs at crossover, surely it too must be crossover distortion, and the original intent is lost?
Have you ever built and compared, both measured and subjective, Class AB amps of varying bias depth and Class A? A few examples might serve your arguments/opinions well. The art is furthered only by correlating topology, measured distortion, and subjective experience, and opinions on the measured performance alone regrettably do not complete the picture. It appears to be much more complex than thought.
BTW, some highly distortive tube circuits subjectively score very highly, even amongst credentialled engineers. Other very low distortion SS amps sound terrible. NP's single ended Class A circuits sound wonderful. Where is the correlation with the measured distortion? Can you explain this to us?
Cheers,
Hugh
AKSA said:
Hmmmm... Perhaps I'm not communicating well
...First, I think listening tests are great. But they're almost useless when trying to compare what you hear in your system, room, etc. to what I hear in mine. So for the purposes of comparing things in an online forum, such as this one, they're of limited value (IMHO).
Second, I have said elsewhere (and I think in this thread?) that what matters most is what the designer/builder/owner/DIYer thinks sounds best. For one person that might be a single ended triode. For another it might be the YAP design.
I agree distortion spectrums are very useful. They help identify distortion vs noise vs power supply hum etc. They also help quantify (i.e. "weight") the more objectionable forms of distortion against the less objectionable kinds.
I've said many times that I think crossover distortion is the most offensive distortion left in most non Class-A amplifiers. So I'm not sure where you got the impression I don't think eliminating it is worthwhile?
I've also said I think it's great to correlate topology, and designs in general, with measured distortion--especially when the measurements are repeatable and comparable online.
As for correlating any or all of the above with *subjective* results, I think that's extremely hard to do. I think the only fair way to do it is with a well run double blind or ABX comparison. And that opens another (usually nasty) can of worms.
When it comes to subjective listening tests, I tend to defer to guys like Sean Olive who know more about it than I do. Douglas Self and many others have written a lot on the subjective issue. You can debate it forever so I try to avoid it. But the short version is what I hear in my listening room with my subjective biases probably won't match what you would hear in yours with your subjective biases.
AKSA,
I missed a few of your points and you may have misunderstood one of mine...
When I suggested "using real music with real speakers" you might have missed what I meant by "differencing" and assumed I meant "listening"? I was talking about subtracting the input signal from the output signal of the amp and evaluating the difference or what's left. This can be done purely in the analog domain (i.e. the Baxendall or Halfer methods) or, with some added benefits, in the digital domain (i.e. Waslo's Recent AES Paper ). You can do spectral analysis on the difference and quantify it in various ways. But, as I said, it's difficult (and not usually fair) to compare difference measurements between different test set ups. But it's very useful as an objective and sensitive means of comparing lots of things under a given set of real world conditions.
When I said "replacing one distortion with another" I was referring to distortions introduced by things like sliding bias circuits or dynamic current injection such as Self uses in his Class XD design. These may reduce distortion at the crossover region, but can introduce new kinds of distortion that were not there before. Self's XD method, for example, time shifts the crossover distortion away from the zero crossing where he claims the resulting new distortion is less objectionable. The distortion curves all change, in some areas for the better, but still don't match a Class A performance overall.
I have built Class B/AB amps of varying bias depths and low power Class A amps. I have not tried to build any "magic bias" amps with the likes of the LT1166, etc. I have more experience with MOSFET outputs than BJTs. And MOSFET's don't exhibit the gm doubling issue. So they generally perform better the more bias you give them. But I have enough experience with BJT amps to verify what Self has documented in reference to gm doubling.
I also have participated in enough blind and ABX testing to know I can't always trust my own subjective (non-blind) listening evaluations. This is especially easy to do with things in the digital domain--such as comparing filter or resampling algorithms. You can use Foobar 2000 to conduct your own ABX of the results. And it's interesting (and humbling) how things you're certain you heard during non-blind listening disappear, or reverse, when you don't know what is what. But this is getting off the topic of bias classes and class G output stages
I still don't know any better way to compare such things on a forum than actual measurements made in a standard way.
I missed a few of your points and you may have misunderstood one of mine...
AKSA said:
When I suggested "using real music with real speakers" you might have missed what I meant by "differencing" and assumed I meant "listening"? I was talking about subtracting the input signal from the output signal of the amp and evaluating the difference or what's left. This can be done purely in the analog domain (i.e. the Baxendall or Halfer methods) or, with some added benefits, in the digital domain (i.e. Waslo's Recent AES Paper ). You can do spectral analysis on the difference and quantify it in various ways. But, as I said, it's difficult (and not usually fair) to compare difference measurements between different test set ups. But it's very useful as an objective and sensitive means of comparing lots of things under a given set of real world conditions.
AKSA said:
When I said "replacing one distortion with another" I was referring to distortions introduced by things like sliding bias circuits or dynamic current injection such as Self uses in his Class XD design. These may reduce distortion at the crossover region, but can introduce new kinds of distortion that were not there before. Self's XD method, for example, time shifts the crossover distortion away from the zero crossing where he claims the resulting new distortion is less objectionable. The distortion curves all change, in some areas for the better, but still don't match a Class A performance overall.
AKSA said:
I have built Class B/AB amps of varying bias depths and low power Class A amps. I have not tried to build any "magic bias" amps with the likes of the LT1166, etc. I have more experience with MOSFET outputs than BJTs. And MOSFET's don't exhibit the gm doubling issue. So they generally perform better the more bias you give them. But I have enough experience with BJT amps to verify what Self has documented in reference to gm doubling.
I also have participated in enough blind and ABX testing to know I can't always trust my own subjective (non-blind) listening evaluations. This is especially easy to do with things in the digital domain--such as comparing filter or resampling algorithms. You can use Foobar 2000 to conduct your own ABX of the results. And it's interesting (and humbling) how things you're certain you heard during non-blind listening disappear, or reverse, when you don't know what is what. But this is getting off the topic of bias classes and class G output stages
I still don't know any better way to compare such things on a forum than actual measurements made in a standard way.
It looks like the debate goes on. A vs. B and the evil Xover distortion . I have heard the big stealth at 200 watts burning through it's 3KVA of toriodial power in class A , my big symasym at the same 200w with it's small amount of crossover "garbage". The difference is so small that the class B's efficiency outweighs the VERY minor class A "superiority".
In my opinion , the fact that we know the B has this distortion might influence us subjectively. We see it in our simulations , we know it adds minor amounts of higher order distortion to the amp , but when you listen ???? ... nothing..
At the first watt or listening to the amps using headphones at the output a very slight "holographic" improvement to the soundstage can be heard .. but you would be strained to discern the differences.
Just to be certain (and fair .. this is DIY) I did these same tests on a full discrete EF-2 sherwood (100w) and a sony HT receiver ... they were DEFINITELY inferior , both in the soundstage and headroom department.
I have built the krill and it worked well. It's "sound" comes from the diamond buffer , not the fact that it "does not switch". Another prototype, "nakamitchi clone" with a standard diamond and EF OP sounded identical.
(The class A is the closest )
OS
. I have heard the big stealth at 200 watts burning through it's 3KVA of toriodial power in class A , my big symasym at the same 200w with it's small amount of crossover "garbage". The difference is so small that the class B's efficiency outweighs the VERY minor class A "superiority".In my opinion , the fact that we know the B has this distortion might influence us subjectively. We see it in our simulations , we know it adds minor amounts of higher order distortion to the amp , but when you listen ????
... nothing..At the first watt or listening to the amps using headphones at the output a very slight "holographic" improvement to the soundstage can be heard .. but you would be strained to discern the differences.
Just to be certain (and fair .. this is DIY) I did these same tests on a full discrete EF-2 sherwood (100w) and a sony HT receiver ... they were DEFINITELY inferior , both in the soundstage and headroom department.
I have built the krill and it worked well. It's "sound" comes from the diamond buffer , not the fact that it "does not switch". Another prototype, "nakamitchi clone" with a standard diamond and EF OP sounded identical.
Absolutely , the same symasym voltage stage driving a EF2 mosfet output stage sounded better than any of the above . I could use half the number of OP devices and overbias to my hearts content. With analog techniques there is no "perfect amp"
(The class A is the closest
)OS
RocketScientist,
I`m much more skeptical about some aspects in Self`s presentation of distortion in output stages.
You must have a hard time matching measurement figures against subjective impression.
I`m much more skeptical about some aspects in Self`s presentation of distortion in output stages.
You must have a hard time matching measurement figures against subjective impression.
Agreed. That`s what error correction techniques, like global feedback, do - can have a favorable outcome as distortions are not equally objectionable.
I find that line muddy.
RocketScientist said:
Hello RS
For the analog domain differencing, there is a web page giving
ideas and practical method.
http://www.angelfire.com/ab3/mjramp/dtest.html
Bye
Gaetan
gaetan8888 said:
Thank you Gaetan8888. I think differencing is a very good method for comparing designs and design changes. It can also be used for small adjustments like bias...
For example, one can monitor the difference signal in realtime on a spectrum analyzer while playing real music into real speakers at typical listening levels as you adjust the bias. This takes into account all the real-world thermal bias distortions that can be very different compared to static bench sinewave analysis.By minimizing the most audible and/or objectionable portions of the difference signal while playing real music into a real load, you can be fairly certain you're also optimizing the sound of the amplifier as long as your differencing method is correct. Although sometimes the trade off of a design change is not clear with differencing (i.e. trading a given amount of midrange distortion for the same amount of high frequency distortion in adjusting loop compensation for example). Then I think when it's unclear it's best to simply listen to both versions in the normal way and/or listen to the differencing signal alone.
One can record the various differencing signals for each design change (I do so digitally via a very high quality ADC and preamp) for later playback and analysis. Of course, if you're using Waslo's digital method, they're already recorded. While the speakers are playing it is difficult, even with headphones, to listen to the difference signal. But by recording it you can play it back and listen with headphones when the system is silent. It's fairly easy then to judge which one is least objectionable. Some distortions are much more "musical" than others.
Digital recording also lets one easily calculate the total energy within the audible band over time which gives an objective measure to use for numerical comparison. The value indicates the total of nearly all kinds of amplifier distortions under the real world test conditions. In the digital domain you can calculate an average over time of an entire song or even a collection of very different kinds of music. I use large FLAC files that each have several different kinds of reference music combined into a single reference file.
I start the digital recording of the difference signal, sit down in the sweet spot, play my reference FLAC file, enjoy the good sound, and when the file is done, both listen to and analyze the difference signal. The resulting file contains everything the amplifier did wrong while playing all the different music.
I can say (also reported by others) how an amplifier performs in a differencing test under real world conditions is not always predicted by how it performs with sine waves into a dummy load. I have made changes that lower the steady state THD but make the difference test results both subjectively and objectively worse. But, as I have said, it's difficult to compare difference results outside of a given test setup. So, for comparing things in a forum between different people, THD and similar repeatable measurements are still best.
Listening to the difference signal, you can also plainly hear the gm doubling and crossover distortion in an overbiased BJT amp. When the music is at a level mostly within the Class A region but occasionally goes higher, you hear (and see on the spectrum or scope) little spikes of noise as the output stage briefly transitions from Class A to Class B.
Bill Waslo has some interesting differencing files, including 3 different power amps, obtained with his software here:
Differencing Examples
If you listen to the difference tracks for the commercial home theater monoblock compared to the DIY modified Hafler you'll hear the Hafler's distortion is both more musical (less harsh) and lower in level. You have to crank up the "playback boost", however, to hear much of anything from either one. The IcePower sounds surprisingly similar to the Hafler because most of its extra distortion is above the audio band.
I've found the results obtained with Waslo's software can be different than you get with an analog null circuit. On the plus side the digital method can eliminate ground loop issues and noise/distortion from the analog differencing circuit. But the "digital null" can produce a different result than an analog null for a variety of reasons.
Ideally, as Bill points out in his AES paper, you want a common sample clock between the input and output signal. Under those conditions, his method yields very good results. Otherwise his software has to try and correct for clock drift. IMHO, digitally recording an amplified analog difference signal offers a good compromise. The analog method also removes objections from digiphobes who try to argue you lose critical details in the A/D process.
RocketScientist said:
Thank you Gaetan8888. I think differencing is a very good method for comparing designs and design changes. It can also be used for small adjustments like bias...
Hello RS
I agree with the use of real speakers for testing, using a dummy load are far from real life tests result.
What scope you use to see the cross-over spike ?
My Sencore SC61 scope do not show them, but it is not a long time that I have it, maby I do not use it the right way.
Thank
Bye
Gaetan
gaetan8888 said:
It depends. Sometimes I use the "scope" built into various PC software applications that I use when analyzing files that are already in the digital domain (i.e. .WAV recordings of the difference signal). But I have an Agilent MSO 6000 series scope I use in real time and also a 100 Mhz Tektronix analog scope.
Set the trigger level of the scope to trigger on the music (not difference) signal only when the music exceeds the Class A region of the amp, With both the music and difference traces displayed, you should see a spike in the difference signal corresponding with the music peak and the switch from Class A to Class B operation.
Otherwise crossover distortion can be hard to see in a real time residual or difference signal without using a sine wave source, triggering the scope on the sinewave, and averaging the result on a digital scope. The averaging removes the random noise and leaves the periodic noise. Because the scope is triggered on the sinewave, the crossover distortion becomes very obvious. This is how most of the classic scope traces you see published are made like this one:
Attachments
RocketScientist,
In my view, the balanced bridge is the most exciting balanced circuit. It cancels all the disturbances balanced circuits cancel, plus some more (as a special bonus), does not offer the purity of single-ended Class A, which is, however, out of consideration for you (aiming at 1600W), so I would not hesitate one minute.
- Klaus,
right?
I'm sure. It means an unbalanced circuit, where the (asymmetric) signal is referred to ground.
In my view, the balanced bridge is the most exciting balanced circuit. It cancels all the disturbances balanced circuits cancel, plus some more (as a special bonus), does not offer the purity of single-ended Class A, which is, however, out of consideration for you (aiming at 1600W), so I would not hesitate one minute.
Nice, but with bipolar output I would attack the AD847 topology.
- Klaus,
right?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.