I'm not going to argue what can be done with local VS global feedback, but I will explain some of my findings.
I prefer keeping the feedback loops as short as possible. I prefer applying local feedback to the output stage as the primary feedback mechanism. This is a "first line of defense" against distortion, simply because that is where most of the distortion is generated.
I designed an amp I call the simple single ended, or SSE for short since lawyers have determined that I can't use the word simple together with the common abbreviation for single ended. It is designed to use the 6V6, 6L6GC, EL34 or KT88 type tubes. It is pretty much like any other two stage amp with two exceptions.
The driver stage (12AT7) is CCS loaded using an IC. A CCS is a perfect load for a triode, and the IC is a near perfect CCS. This reduces driver distortion to a very low level and raises the stage gain to near Mu.
The output stage can be jumper (or switch) selectable for triode, UL, or pentode operation. The full speaker output can be applied to the cathode of the output tube if desired (switch or jumper selectable). Depending on output tube and OPT choice this provides between 5 to 8 db of local feedback.
A few hundred of the PC boards for this design have been sold and builders have made amps in every possible configuration I could think of, and some I never thought of. User feedback reveals the popular combinations are the triode strapped 6V6 or EL34 with zero feedback for mostly simple music that the SET crowd favors, and the KT88 or 6550 in UL with cathode feedback for louder more complex music. These are indeed my two favorites.
The simple cathode applied local feedback will extend the measurable low frequency response of a budget OPT like the $29 Edcor or the Hammond 125CSE. It makes a very obvious difference in the sound quality with these OPT's, but does less for a good transformer.
Agreed, you can only go so far with cathode feedback. You can however feedback some of the signal on the plate directly into the grid of the output tube. If the driver tube is a high Rp pentode a simple resistor from the output tube's plate to the driver tube's plate is all that's needed.
I prefer a summing junction whose output is buffered by a mosfet follower before direct coupling to the output tube's grid avoiding the coupling cap's overload recovery issues and allowing for DC bias adjustment.
Feedback applied in this manner can be increased without bounds until the output tube has zero gain, but that is not needed or wanted.....I would rather use a cathode follower output stage....done that too.
I apply enough local plate to grid feedback to reduce the output impedance and clean up THD while watching the higher order harmonics. Stop when they no longer drop, since some 2H and 3H are relatively harmless.
In a 3 stage P-P amp it helps to apply a small amount of cross coupled feedback from the output tube's plate to the plate of the input stage. This reduces odd order harmonics and crossover distortion.
IF the OPT is symmetrical enough some balanced feedback from the OPT secondary (ground the 4 ohm tap, FB from 0 ohm and 16 ohm tap) will work too. I usually apply this to the cathode of the output stage, or the driver cathodes. If the OPT has good symmetry, the higher order odd's will drop, if not, they may get worse. Toroidal OPT's even some repurposed mains toroids work good here.
There is no single "magic bullet" for feedback schemes, so all options must be tried with measuring equipment and static tones, and verified with music on real speakers.
I am working on methods to measure, or at least observe distortion with music and speakers, but am not there yet.
If you have a two channel digital storage scope with the ability to invert and subtract the two channels, you can try this. My 20 year old TEK2232 works fine.
Hook a source to the amp's input that can produce tones or music. I use a computer and a 24 bit 192 KHz sound card, but a CD player and a test CD will work.
Put one scope channel on the amp input, and one on the speaker output, and load the amp with a pure restive load. Apply a 1 KHz tone at a level to drive the amp hard, but not into clipping. Adjust the scope to display each signal one on top of the other filling most of the screen. Press the invert button, then the subtract button. You should now have a flat line that shows only the amp's distortion. Crank the input until the amp clips, and you will see the distortion increase.
Apply some music, and as long as you stay out of clipping, you SHOULD see nothing.....HMMMM what is all of that stuff....hopefully it is still at a very low level.
Reduce the input, remove the load, connect the speakers, and repeat. Gee, the amp behaves different with a speaker on it. Yes, an SE amp with it's high output impedance is the worse case and a class D chip amp is still somewhat behaved at least with an 8 ohm small speaker. Each amp/speaker combination behaves differently. Test with highly dynamic music. Flamenco guitar, or Latin drums at high volume really makes my Yamaha NS-10M studio speakers misbehave when fed with an SE amp.
I am now working with a multi channel recording setup with taps all over the amp to try and quantify what is happening, but I believe that some speakers can generate enough counter EMF on a drum hit during a bass guitar note to make their instantaneous impedance appear negative. This has been debated or, disbelieved by members of this forum, but until I have more data, I can't say for sure.
I prefer keeping the feedback loops as short as possible. I prefer applying local feedback to the output stage as the primary feedback mechanism. This is a "first line of defense" against distortion, simply because that is where most of the distortion is generated.
I designed an amp I call the simple single ended, or SSE for short since lawyers have determined that I can't use the word simple together with the common abbreviation for single ended. It is designed to use the 6V6, 6L6GC, EL34 or KT88 type tubes. It is pretty much like any other two stage amp with two exceptions.
The driver stage (12AT7) is CCS loaded using an IC. A CCS is a perfect load for a triode, and the IC is a near perfect CCS. This reduces driver distortion to a very low level and raises the stage gain to near Mu.
The output stage can be jumper (or switch) selectable for triode, UL, or pentode operation. The full speaker output can be applied to the cathode of the output tube if desired (switch or jumper selectable). Depending on output tube and OPT choice this provides between 5 to 8 db of local feedback.
A few hundred of the PC boards for this design have been sold and builders have made amps in every possible configuration I could think of, and some I never thought of. User feedback reveals the popular combinations are the triode strapped 6V6 or EL34 with zero feedback for mostly simple music that the SET crowd favors, and the KT88 or 6550 in UL with cathode feedback for louder more complex music. These are indeed my two favorites.
The simple cathode applied local feedback will extend the measurable low frequency response of a budget OPT like the $29 Edcor or the Hammond 125CSE. It makes a very obvious difference in the sound quality with these OPT's, but does less for a good transformer.
Agreed, you can only go so far with cathode feedback. You can however feedback some of the signal on the plate directly into the grid of the output tube. If the driver tube is a high Rp pentode a simple resistor from the output tube's plate to the driver tube's plate is all that's needed.
I prefer a summing junction whose output is buffered by a mosfet follower before direct coupling to the output tube's grid avoiding the coupling cap's overload recovery issues and allowing for DC bias adjustment.
Feedback applied in this manner can be increased without bounds until the output tube has zero gain, but that is not needed or wanted.....I would rather use a cathode follower output stage....done that too.
I apply enough local plate to grid feedback to reduce the output impedance and clean up THD while watching the higher order harmonics. Stop when they no longer drop, since some 2H and 3H are relatively harmless.
In a 3 stage P-P amp it helps to apply a small amount of cross coupled feedback from the output tube's plate to the plate of the input stage. This reduces odd order harmonics and crossover distortion.
IF the OPT is symmetrical enough some balanced feedback from the OPT secondary (ground the 4 ohm tap, FB from 0 ohm and 16 ohm tap) will work too. I usually apply this to the cathode of the output stage, or the driver cathodes. If the OPT has good symmetry, the higher order odd's will drop, if not, they may get worse. Toroidal OPT's even some repurposed mains toroids work good here.
There is no single "magic bullet" for feedback schemes, so all options must be tried with measuring equipment and static tones, and verified with music on real speakers.
I am working on methods to measure, or at least observe distortion with music and speakers, but am not there yet.
If you have a two channel digital storage scope with the ability to invert and subtract the two channels, you can try this. My 20 year old TEK2232 works fine.
Hook a source to the amp's input that can produce tones or music. I use a computer and a 24 bit 192 KHz sound card, but a CD player and a test CD will work.
Put one scope channel on the amp input, and one on the speaker output, and load the amp with a pure restive load. Apply a 1 KHz tone at a level to drive the amp hard, but not into clipping. Adjust the scope to display each signal one on top of the other filling most of the screen. Press the invert button, then the subtract button. You should now have a flat line that shows only the amp's distortion. Crank the input until the amp clips, and you will see the distortion increase.
Apply some music, and as long as you stay out of clipping, you SHOULD see nothing.....HMMMM what is all of that stuff....hopefully it is still at a very low level.
Reduce the input, remove the load, connect the speakers, and repeat. Gee, the amp behaves different with a speaker on it. Yes, an SE amp with it's high output impedance is the worse case and a class D chip amp is still somewhat behaved at least with an 8 ohm small speaker. Each amp/speaker combination behaves differently. Test with highly dynamic music. Flamenco guitar, or Latin drums at high volume really makes my Yamaha NS-10M studio speakers misbehave when fed with an SE amp.
I am now working with a multi channel recording setup with taps all over the amp to try and quantify what is happening, but I believe that some speakers can generate enough counter EMF on a drum hit during a bass guitar note to make their instantaneous impedance appear negative. This has been debated or, disbelieved by members of this forum, but until I have more data, I can't say for sure.
Something I haven't seen mentioned here so far is that in Douglas Self's book, in addition to the Blameless concept he also gives a list of amplifier distortion mechanisms and how to reduce them. I'm guessing there is a similar list of tube amplifier distortions along with ways to minimize them that can be made. Maybe something like "Output Transformer Distortion", or "Phase Inverter Distortion" for example.
Exactly. That's what I've been saying. Self separated out and understood each distortion mechanism in each stage. Poeple seem instead to just think he designed a solid state amplifier.
When Self published his work in the 1990's, he did something that had never been done before for that solid state topology. Top engineers in Quad, Matshushita most likely had as good an understanding, but they didn't publish a comprehensive work on the subject.
I also said very early in this discussion that I had considered doing the same for one or more tube amp topologies, but decided not to, because:-
a) Self's work went unappreciated by many
b) there's no money in doing the same for tube amps
c) with respect to tube amps, it has all (that is, understanding the distortions in each stage & the transformer) been done before.
Allmost any good tubemakers' handbook tells you how to calculate the distortion caused by triode and pentode voltage amplifying stages and simple class a power stages. The behaviour of transformers has been fully analysed in multitudes of transformer textbooks. I have 1950's application lab reports on ultrlinear. Other distortion sources, such as caused by bias shift in power stages having bypassed cathode resistors is a bit more obscure, but papers on it are out there.
Tubes were the only engineered audio technology for nearly 45 years. Thousands of eminent factory engineers and university researchers were involved. Do you really think they didn't know what they were doing?
Last edited:
I assume you mean they have advised you that registering "simple SE" as a trade mark is open to legal challenge, which it certainly would be, not that you can't use it.
Otherwise, trademark law in the USA must be strange, or you need to choose a better lawyer.
The perfect load for a triode is another triode. The so-called self balancing push-pull circuit (SB-PP) causes the upper triode (the load triode) to automatically adjust its' instantaneous resistance to match teh opposite change in the lower (amplifying) triode. A CCS load is merely distorion reducing. The SB-PP circuit balances out distortion.
To get perfect cancellation the two triodes you need matched triodes, but it isn't critical and twin triodes come pretty well matched.
The only problem with the SB-PP is the upper triode, because its' cathode is not earthed, can introduce hum. That's seldom a problem in power amp stages though, asuming you choose an appropriate twin triode.
When you say 'cross-coupled' do you mean from the 'push' part back to the 'pull' and vice versa like this?
An externally hosted image should be here but it was not working when we last tested it.
Does this just work on class A or is the 'off' primary/anode driven from the transformer anyway and still gives the correct(ish) feedback?
The preponderance of design mediocrity suggests not all of them did
Many of the theories used to design amplifiers are old but incomplete and simplistic. The two main ones that spring to mind are:1) PSRR
Look at a textbook design of (e.g) a common cathode stage, it's PSRR will be stated as 6dB. That's wrong however unless you have no signal because it's a simplistic theory - what actually happens is that positive outgoing signal gets lots of PSU line noise and negative going signal gets almost none - i.e. the PSRR is signal dependent and variable. PSRR also includes a non-linear imprint of the music you are playing too - i.e. it's dynamic and changes - any SPICE simulation using a perfect B+ rail has IMO already failed.
2) Negative feedback
The simplistic failure here is that the error correction is assumed to pass through a linear amplifier on it's travels back to the output, whereas in fact it passes through exactly the same non-linear one you are trying to correct.
True believers in GNFB don't understand that their pet theory only really works on a linear amplifier - i.e. one that doesn't need correcting (barring side effects like lowering impedance).
It took the Japanese to realise that GNFB was just moving problems - not solving them.
Unless we look at a music signal as it travels through the amplifier (and stop fixating on a low-signal 1kHz tone) we'll miss the subtle mechanisms that mangle the signal on the way to the air.
To a large extent this means a short path between input and speaker terminals - GNFB multiplies that short path (by definition) into a variable set of competing amplifiers at various levels and phases and we listen to them all - again it took the Japanese SET movement to realise that listening to just one one amplifier, once, was better.
And goodness knows why you'd want to listen to your OPT more than once lol
The preponderance of design mediocrity suggests not all of them did
Just because you can find some rubbish, whether textbooks, learned papers, or actual hardware, means nothing. The worst available is by definition the worst.
As I used to be in the radio & TV service business in the 1960's I well recall those horrible German el cheapo record players - I don't know if they were imported into the UK. They had a tinplate case (hopeless for acoustics and durability), a selenium rectifier (common German practice and ok), and a single pentode. The case was held together by bent tabs (no screws), just like tinplate toy model cars at that time.
Does this mean that the Germans only knew how to make rubbish? Of course not. Reputable German firms knew how to make precision high quality products and had a reputation for it.
Before tubes they appears to also make them too!
Tennants Auctioneers: A German Clockwork Tinplate 'Lemiphone' Child's Record Player
£300 for that one - we should have been stocking up
No, the most of us understood that "blameless" is not an amplifier, but a standard of excellence, and a series of methods to achieve it, along with a profound analysis of how and why.
No. It seems that another company has already registered that exact combination of letters in that order without the space for use with a solid state home theater box. That company has been using a high powered legal firm that has "52 excellent lawyers on staff."
Long ago before said high powered legal firm was involved I had received an email from one of their vice presidents who had discovered my "misuse of their trademark." We both agreed that we were not in any way competing, and my use was OK. I still have that email.
Several years later it seems that a Google search for their product by name brings up dozens of pictures on proud SSE builders displaying their creations. The HT product wasn't even in the top 10 hits. Said large company is justifiably unhappy, so said large legal firm sends me a "cease and desist all violations" letter.
I talked to a lawyer who advised me that they "squash me like bug on their corporate windshield" if they wanted to, so I agreed to not use the name, change the website, but continue to ship the stock of boards until depletion.
Exactly. Several experts here will tell you that you can't insert feedback via a resistor into the plate of a triode as shown here due to the varying plate resistance. However it does work quite well in LTP situations because of the high output impedance. Use a CCS in the tail for best results.
As we all know "perfect" only exists in textbooks, and the "best" load for a triode in the real world depends on what it is driving. (AC load impedance). The 12AT7 is not known for great linearity, especially with a resistor load. I needed a common tube with world wide availability for the SSE, and it had to have enough gain to drive a KT88 to clipping in a two stage "simple" amp. Testing with several hundred tubes of about 10 common types, revealed that a 12AT7 with a CCS load was the best choice for a simple design. The PSRR afforded by the CCS is excellent too.
In more complex designs I CCS load a high Gm triode, and buffer its output with a mosfet follower, thus presenting the triode with a near infinite load at DC and AC. Measured results show near zero distortion with a high Gm triode like the 5842. This method is also useful when large driving voltage (hundreds of volts P-P) is needed for tubes like the 845, screen driven sweep tubes, cathode followers, or large amounts of local feedback. A 6SN7 or a 45 can generate 300+ volts P-P from a 500 volt supply.
Google finds several circuits by this name, but I am assuming you are referring to the SRPP? I have tested that design, and indeed used it in my 300B push pull amp. If you build one and test it with different loads, there is indeed a sweet spot where you will get a good bit of output swing with a low distortion. The distortion rises on either side of this sweet spot.
.
I have a electrical engineering textbook from MIT that was published in 1929. There was already a lot of serious math thrown at the subject long before any serious effort was done to design a low distortion audio amplifier.
What? I've never seen any textbook quote PSRR as anything other than a mathematical function of the components in the circuit, which is perfectly correct.
Last edited:
What the topic starter mentioned is definitely an excellent suggestion. Douglas Self certainly has made a comprehensive summary of all the solid state amplifier distortion mechanisms for the Miller approach. His work is pretty much the defacto standard when it comes to designing a low-distortion solid-state amplifier.
Having something similar for a tube amplifier would most definitely be appreciated by many and in my opinion would quickly become the gold standard for DIY. Certainly if different topologies (PP, SE, etc.) are discussed and evaluated.
As mentioned in this topic aready, there's a wealth of information about tubes, which has been accumulated by engineers, scientists, students, etc. etc. for over 70 years. Everything has already been researched and documented, it just needs to be combined into one easy to read book like Douglas Self's Power Amplifier Design Handbook.
Anybody willing to meet this challenge head on?
Having something similar for a tube amplifier would most definitely be appreciated by many and in my opinion would quickly become the gold standard for DIY. Certainly if different topologies (PP, SE, etc.) are discussed and evaluated.
As mentioned in this topic aready, there's a wealth of information about tubes, which has been accumulated by engineers, scientists, students, etc. etc. for over 70 years. Everything has already been researched and documented, it just needs to be combined into one easy to read book like Douglas Self's Power Amplifier Design Handbook.
Anybody willing to meet this challenge head on?
Ah! they registered first. Bad luck.
Re self balacing push pull, you are correct. Upon googling just now, I too discovered all manner of things not what I meant, but in the (1960's and some 1990's) textbooks and journals I have, they refer to it as Self Balancing Push Pull what now seems to be on the www referred to as SRPP (Shunt Regulated Push Pull).
Yes, there is in practice a sweet spot of load resistance where distortion is minimum. The sweet spot can be deepened and broadened (ie more tolerant of non-optimum loads) if you take the output from the centre point of two resistors used in series for the cathode resistor of the upper triode.
I might if there was compelling evidence that a significant number of people wanted it and would appreciate it. It would have to be published in Linear Audio (if Mr Didden does indeed pay his authors) or Elektor or AudioXpress, so that I can recover at least some of the expense in procuring tubes and transformers. And they might not want it.
Convince me that the interest is there. Then if I don't take it up, someone else will....
Even though it has all been done before, it would still be a considerable amount of work - perhaps 1000's of hours. Practical testing is very expensive, and more important with tubes than it is with bipolar transistors. With transistors you can rely a lot more on SPICE.
Self only did one topology. Paradoxically, I think for tubes at least 3 major topologies should be tackled, perhaps more.
In terms of saleability, I think it would be bettter to do a couple of impotant solid state topologies and add to Self's work on his chosen topology, i.e., a) topologies using power FET's, b) topologies more suited for Class A, and c) switch mode topologies not using specialised IC's.
@Keit,
I think a good starting point would be to just cover the basics of SE and PP amplifier design and look at distortion mechanisms as Douglas has in his book. You can then quickly zero in on topologies that yield the best results. Then follow up with some examples, which illustrate in the simulator how these distortion mechanisms can be tackled and round that off with a practical design that the end-user can build.
Obviously there's many flavors of SE and PP, but for the practical design I would certainly focus on easy to procure parts that yield good results. For example for a SE amplifer a 300B would be a good starting point, for PP a KT88 is a very popular choice. Having said that, the distortion mechanisms are generally universally applicable and would obviously not be limited to the tubes being used.
For the record I am the editor in chief of the Elektor Audio Specials, so I can certainly see whether there's some budget available for a book covering all these aspects. It will however be a significant effort on your (or someone else's) part.
I think a good starting point would be to just cover the basics of SE and PP amplifier design and look at distortion mechanisms as Douglas has in his book. You can then quickly zero in on topologies that yield the best results. Then follow up with some examples, which illustrate in the simulator how these distortion mechanisms can be tackled and round that off with a practical design that the end-user can build.
Obviously there's many flavors of SE and PP, but for the practical design I would certainly focus on easy to procure parts that yield good results. For example for a SE amplifer a 300B would be a good starting point, for PP a KT88 is a very popular choice. Having said that, the distortion mechanisms are generally universally applicable and would obviously not be limited to the tubes being used.
For the record I am the editor in chief of the Elektor Audio Specials, so I can certainly see whether there's some budget available for a book covering all these aspects. It will however be a significant effort on your (or someone else's) part.
I will see what turns up here, and perhaps contact you at Elektor via email.
There is merit in perhaps setting budget for the construction cost to readers to replicate a topology, which is to a certain extent contrary to Doug's approach. In SS design, a bipolar transistor is a bipolar transistor and its internal design and external operating conditions determine what it does. Hence for a good topology the result you get reaches a limit once you have achieved "blameless" design - that is made no design errors. With a tube amp, it's best possible performance is limitted by the "quality" of the transformer - which is determined in a limit by the money spent on it. Which is almost saying there's no limit.
Worse (from the point of view of structuring any textbook or article series), if money is to be limitted in regard to the transformer, then one would choose a different topology than you would if an expensive transformer is used. It has long been established that the lowest distortion achievable when cost is no object comes with ultra-linear with push pull drivers. But the Quad-like arrangment of Neville Thiele is very nearly as good with a MUCH cheaper transformer. Then there is the Murray topology, which achieves 0.02% THD with no global feedback. It was devised at the tail end of the tube era, so few have heard of it, and very few have objectively studied it.
There's also the question of what sort of tubes to use. There are essentially 3 kinds of importance:-
a) Classic design tubes that are fashionable and still made - eg 300B.
Despite what some think (and I can hear the knives coming out) these tubes are more a means of sucking money out of you than genuine high performance
b) Late tube-era tubes specifically intended for audio eg EL34/6CA7 and KT66/77/88
These tubes do a good job but the price has been driven up by the audiophile market and the musical instrument market
c) NOS television scanning tubes.
These were built to high standards and so are reliable and durable, they are cheap, and they give a very good account of themselves in audio service.
Last edited:
@Keit,
Of course the cost of a transformer is a determining factor in a design, but regardless global feedback should be applied to any amplifier if one wants to follow the input signal closely. The goal of such a book should not to be working around a design using a specific transformer type, but a general set of guidelines on how to tackle distortion mechanisms in tube amplifiers, these will be universally applicable.
That's the objective here, at least in my humble opinion. Identifying distortion mechanisms in tube amplifiers and address them to ultimately end up with a set of guidelines on how to design a tube amplifier that offers accurate amplification of the input signal.
Of course the cost of a transformer is a determining factor in a design, but regardless global feedback should be applied to any amplifier if one wants to follow the input signal closely. The goal of such a book should not to be working around a design using a specific transformer type, but a general set of guidelines on how to tackle distortion mechanisms in tube amplifiers, these will be universally applicable.
That's the objective here, at least in my humble opinion. Identifying distortion mechanisms in tube amplifiers and address them to ultimately end up with a set of guidelines on how to design a tube amplifier that offers accurate amplification of the input signal.
I will restate my belief that Williamson, Hafler and Keroes already did for tube amplifiers in the 50s what Self did for solid-state. Besides the UL thing, Hafler and Keroes addressed the Williamson's Achilles heel, its low frequency stability issues.
The only place to go after the Williamson is to improve the recovery from clipping. This is why I mentioned the Crystal Palace, as MJ made clean recovery from overload a design priority, and did a very good job of it IMO. The driver section of this amp could easily be applied to an output stage of less epic proportions.
Having said that, Self didn't care about clipping and his darlington VAS saturates horribly, so maybe recovery from overload is not a criterion for Blamelessness. Certainly clipping was not one of the "Eight Distortions".
The only place to go after the Williamson is to improve the recovery from clipping. This is why I mentioned the Crystal Palace, as MJ made clean recovery from overload a design priority, and did a very good job of it IMO. The driver section of this amp could easily be applied to an output stage of less epic proportions.
Having said that, Self didn't care about clipping and his darlington VAS saturates horribly, so maybe recovery from overload is not a criterion for Blamelessness. Certainly clipping was not one of the "Eight Distortions".
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.