http://www.altmann.haan.de/splif_page/#mod directions
Any comments on this design?
the author claims the following
"The SPLIF (split feedback) topology has all the benefits of global negative feedback.
Like good linearity, very low distortion (or the distortion you like) and as much power, as you desire.
The SPLIF (split feedback) topology has none of the drawbacks of global negative feedback.
It does not try to control the speaker. It does not even notice, that such thing as a speaker exists.
Stability problems from reactive loads are not an issue with the SPLIF amplifier topology.
Yes it seems to be too good to be true, but there is even more to come:
By connecting the outputs of the output stages (4&8) with a switch, you can switch feedback on and off.
So you will actually be able to actually hear, what feedback does.
You can do it yourself with your existing amp.
"
Any comments on this design?
the author claims the following
"The SPLIF (split feedback) topology has all the benefits of global negative feedback.
Like good linearity, very low distortion (or the distortion you like) and as much power, as you desire.
The SPLIF (split feedback) topology has none of the drawbacks of global negative feedback.
It does not try to control the speaker. It does not even notice, that such thing as a speaker exists.
Stability problems from reactive loads are not an issue with the SPLIF amplifier topology.
Yes it seems to be too good to be true, but there is even more to come:
By connecting the outputs of the output stages (4&8) with a switch, you can switch feedback on and off.
So you will actually be able to actually hear, what feedback does.
You can do it yourself with your existing amp.
"
Rubbish. The reactive load currents are reflected back through the stages and still appear on the driver or pre-driver outputs.Nordic said:
What this technique achieves is a reduction in speaker induced feedback to the inverting input node of the amp and removes one very linear stage (output follower) from the Global feedback loop. But it is still a global feedback amplifier (it includes the three stages of a Lin amp topology inside the NFB loop).
lumanauw wrote:
In principle, feed-forward can "fix" because it cancels the output error by subtracting it from the output. There is no hunting for a stable, minimum error equilibrium like there is with feedback. There are no stability considerations.
In terms of effectiveness in a theoretical sense I would say pre-correction is best (IOW making the OL circuit as linear as possible - called pre-distortion in the telecoms world), then post-correction (feed-forward) and lastly degenerative correction (feedback).
Feed-forward is obviously very hard in a power amp because the output subtractor has to be linear at high powers...which is usually impractical. The Quad405 had a power limitation at high frequencies for this reason and its performance was sensitive to speaker make.
Hi. I'm not very expert on the Quad405 but my understanding is that it uses a feed-forward method at high frequencies. I think it uses feedback at low frequencies but I can't recall.
In principle, feed-forward can "fix" because it cancels the output error by subtracting it from the output. There is no hunting for a stable, minimum error equilibrium like there is with feedback. There are no stability considerations.
In terms of effectiveness in a theoretical sense I would say pre-correction is best (IOW making the OL circuit as linear as possible - called pre-distortion in the telecoms world), then post-correction (feed-forward) and lastly degenerative correction (feedback).
Feed-forward is obviously very hard in a power amp because the output subtractor has to be linear at high powers...which is usually impractical. The Quad405 had a power limitation at high frequencies for this reason and its performance was sensitive to speaker make.
Nordic wrote:
Heed these words. Too many designers, both amateur and professional, make the mistake of choosing a popular measurable parameter or two and, assuming the sound quality is strongly correlated to them, they build circuits which optimise them. They often don't perform well sonically. The thing I respect about Altmann is that he chose to believe that the humble SET sounded great and then tried to figure out why the transistor amps he heard didn't.
It's a clever idea. He has sort of made two output stages in parallel, one feeding the speaker and the other a dummy load. By taking the FB from the dummy output he corrects some of the distortion of the speaker output while mostly isolating the FB from the impedance effects of the speaker. It's a half-way house between GNFB and no FB.
Like Andrew I don't think his claims are water-tight but I can believe that his circuit sounds much better with the switch open.
"The SPLIF (split feedback) topology has all the benefits of global negative feedback."
It doesn't have the benefit of sensing the distortion at the speaker output...so there is distortion that the FB system cannot reduce.
"Like good linearity, very low distortion (or the distortion you like) and as much power, as you desire."
Sure.
"The SPLIF (split feedback) topology has none of the drawbacks of global negative feedback."
Not sure that it has none of the drawbacks. It depends what you recognize as NFB drawbacks. It's not as global but is still partly global as Andrew points out.
"It does not try to control the speaker. It does not even notice, that such thing as a speaker exists."
See above.
"Stability problems from reactive loads are not an issue with the SPLIF amplifier topology."
See above. Less of an issue, sure.
Charles Altmann wrote:
Heed these words. Too many designers, both amateur and professional, make the mistake of choosing a popular measurable parameter or two and, assuming the sound quality is strongly correlated to them, they build circuits which optimise them. They often don't perform well sonically. The thing I respect about Altmann is that he chose to believe that the humble SET sounded great and then tried to figure out why the transistor amps he heard didn't.
It's a clever idea. He has sort of made two output stages in parallel, one feeding the speaker and the other a dummy load. By taking the FB from the dummy output he corrects some of the distortion of the speaker output while mostly isolating the FB from the impedance effects of the speaker. It's a half-way house between GNFB and no FB.
Like Andrew I don't think his claims are water-tight but I can believe that his circuit sounds much better with the switch open.
"The SPLIF (split feedback) topology has all the benefits of global negative feedback."
It doesn't have the benefit of sensing the distortion at the speaker output...so there is distortion that the FB system cannot reduce.
"Like good linearity, very low distortion (or the distortion you like) and as much power, as you desire."
Sure.
"The SPLIF (split feedback) topology has none of the drawbacks of global negative feedback."
Not sure that it has none of the drawbacks. It depends what you recognize as NFB drawbacks. It's not as global but is still partly global as Andrew points out.
"It does not try to control the speaker. It does not even notice, that such thing as a speaker exists."
See above.
"Stability problems from reactive loads are not an issue with the SPLIF amplifier topology."
See above. Less of an issue, sure.
traderbam said:
And many others (reviewers, designers, etc), and likely a great deal more, have listened also, and they do not conclude with or agree with nebulous assertions such as that NFB “sucks the life out of the music” – and talk about a misleading statement.
A well designed solid state amplifier utilising NFB and generating very low amounts of distortion will not “suck the life out of music” – it simply won’t add anything to it. This IMHO, is what an amplifier intended for accurate “HiFi” reproduction is supposed to do.
Some people prefer amplifiers that generate copious amounts THD and IMD (such as the open-loop SET brigade) which add distinctly audible colourations.
Some don’t.
I don’t mind my various valve amplifiers for amplifying the outputs of my various WWII era valve radios, or vintage recordings on my reel-reel units from the 60's and 70's, which are hardly “HiFi” to begin with, but I sure wouldn’t prefer one over a decent ultra linear solid state design for serious amplification.
traderbam said:
Yeah, and the primary way to do that is ensure that the system isn’t grossly non-linear (or lacking in a few other important departments) before applying feedback. This had been known for decades already and is theoretically backed with well-established science. There are very few mysteries to reliable and sound solid state and “hollow state” design with NFB, despite the convictions and utterances of some.
Glen, did you get out of bed on the wrong side this morning? 😉
My statement was not meant to be misleading. It is my experience.
I didn't say that a "properly designed" NFB amp will suck the life out of the music. But, in my experience, these are few and far between. I have seen many designers going to a lot of trouble to reduce or eliminate GNFB because they have found it to degrade the sound in their circuits. Eg: reputable designers like Nelson Pass and John Curl.
My statement was not meant to be misleading. It is my experience.
I didn't say that a "properly designed" NFB amp will suck the life out of the music. But, in my experience, these are few and far between. I have seen many designers going to a lot of trouble to reduce or eliminate GNFB because they have found it to degrade the sound in their circuits. Eg: reputable designers like Nelson Pass and John Curl.
I haven't seen much that I would call "well-established science".
traderbam said:
Nope, same as usual.
traderbam said:
Right. So your hobby seems to be listening to ill-designed amplifiers and determining that they demonstrate something worthwhile about the intrinsic value of NFB.
traderbam said:
Do you actually realise what logically concludes from this declaration, following your "properly designed" comment?
I'd say that they have a personal taste for a certain type and degree of coloration, but you are entitled to your own opinion.
traderbam said:
Right. It's all voodo, revelation and good luck.
Okay....
I sense you want to make a point about NFB not being bad as an idea in itself. Sure.
Erm...I don't think I said that exactly. But the usual reason for applying NFB is to improve an amp that has unacceptable performance. So I'm not sure NFB has any intrinsic value in this context unless it can improve an "ill-designed" amplifier.
I sense you want to make a point about NFB not being bad as an idea in itself. Sure.
traderbam said:
The 3rd harmonics at 60 and 69 KHz are 1/9 the amplitude of the fundamentals, the modulation products at 17, 26, 63, and 66 KHz are 1/3 the amplitude of the fundamental, and 180 degrees out of phase, all for a gain of little over 6dB. This isn't a problem?
Easy to prove with some trig identities.
Mathematical proof of this and a more general finding to come later.
I hope everybody noticed that in Brian's example, there's no linear term in the input-output relationship of the open-loop device at all - only a cubed term. The only reason there are any fundamental components at the output in the non-feedback case at all is because of the identity:
cos3(x) = 1/4(3 cos(x) + cos(3x))
So the open-loop device might be better named a frequency multiplier than any sort of amplification device.
cos3(x) = 1/4(3 cos(x) + cos(3x))
So the open-loop device might be better named a frequency multiplier than any sort of amplification device.
G.Kleinschmidt said:
No question many design variations on the Aphex Aural Exciter but that's a mis-direction and beside the point. The argument is ironically self-contradictory. It's not that difficult to design no-GNFB SE tube circuits that meet every scientific criteria I know for inaudible distortion, noise and frequency response deviation so why would you discount them on the basis of topology while arguing it's invalid to discount GNFB circuits on the basis of topology?
Nordic said:
+ Stark monaural poweramps
(Audiolabor, 1982, designer Reinhard Wachowiak, 200W/8ohm, 3x Sanken 2SA1215/2SC2921 output stage )
By coincidence, I proposed a "SPLIF"-like topology the other day here. I think with a good (that is, linear -- e.g. by means of local FB) output stage this is an worthwhile approach. It's maybe somewhat irrational but I'm quite attracted by the idea of seperating FB loops in terms of the signal domains they handle -- one for voltage (global FB), one for current (local FB in the OS). OTOH there are good designs around which do just the opposite, making the error signal for the global loop *very* output current dependent (but in a very controlled way, by using a OS with a high but nevertheless stable output impedance).
Mr. Altmann is a quite controversal figure in the scene, I'd say. I agree with Andrew that his conclusions on SPLIF properties are not all correct, especially the stability margin stuff with odd loads. It is an as global loop as it was before, it only "seperates out" the load current from the GFB. Not sure what it does WRT crossover distortion... with class A I feel it's working well.
Yet there is one thing I can fully agree upon with him: Many layers of hard laquer definitly kill the sound of an electric guitar or bass....
- Klaus
Mr. Altmann is a quite controversal figure in the scene, I'd say. I agree with Andrew that his conclusions on SPLIF properties are not all correct, especially the stability margin stuff with odd loads. It is an as global loop as it was before, it only "seperates out" the load current from the GFB. Not sure what it does WRT crossover distortion... with class A I feel it's working well.
Yet there is one thing I can fully agree upon with him: Many layers of hard laquer definitly kill the sound of an electric guitar or bass....
- Klaus
Andy wrote:
I chose a simple non-linear OL gain to make the point about FB creating new spectral content as clear as possible.
Do you agree that I could have used y = x + ax^2 + bx^3 + ... instead and shown the same effect?
I would argue it is still an amplifier, pretty distorted though. I think I've heard one or two of these in my time.
I chose a simple non-linear OL gain to make the point about FB creating new spectral content as clear as possible.Do you agree that I could have used y = x + ax^2 + bx^3 + ... instead and shown the same effect?
traderbam said:
Or: y = (x*0.5 + 0.5)^2 - 0.25, which represents a valid transfer function. (not rectifying the signal)
- Status
- Not open for further replies.