Hi Myhrrhleine,
At first sight, I also couldn't believe this. Indeed, the (total!) THD decreases, BUT the spectral content has changed. Suppose we have an amp (without FB) which produces only 3rd harmonic distortion. By applying FB we put the 1st harmonic AND the 3rd harmonic from the output back to the (inverting) input. Now, this FB signal is also distorted by the amplifier, resulting in 1st, 3rd and 3*3=9th harmonic components. Admittedly, the latter is generally very small.
Regards, Edmond.
PS: Courtesy of Bob Cordell, who made this clear to me.
Hi Edmond,
There's a really great article by Baxandall in Wireless World where he goes into a lot of detail (both mathematical and measurements) about the effect of varying levels of feedback on the different harmonic orders. You're welcome to shoot me an email if you'd like a copy. Anybody else who wants one is welcome to do so as well.
There's a really great article by Baxandall in Wireless World where he goes into a lot of detail (both mathematical and measurements) about the effect of varying levels of feedback on the different harmonic orders. You're welcome to shoot me an email if you'd like a copy. Anybody else who wants one is welcome to do so as well.
Andy,
I think we talk about the same graph. I may have mixed up Peter Baxandall with John LH on this one.
One thing to put it in perspective: the original graph assumes an amp stage with a quadratic transfer function. If the tf is different the results are likely to be different as well, but I would expect a similar effect, as explained by Edmond as well.
Jan Didden
I think we talk about the same graph. I may have mixed up Peter Baxandall with John LH on this one.
One thing to put it in perspective: the original graph assumes an amp stage with a quadratic transfer function. If the tf is different the results are likely to be different as well, but I would expect a similar effect, as explained by Edmond as well.
Jan Didden
Jan,
The Baxandall article is cool because he does measurements to verify the theoretical graph, using as the open-loop amp a FET circuit whose distortion is dominated by second-order, but has a small amount of higher-order distortion components. His measurements are pretty close to the theoretical graph.
The Baxandall article is cool because he does measurements to verify the theoretical graph, using as the open-loop amp a FET circuit whose distortion is dominated by second-order, but has a small amount of higher-order distortion components. His measurements are pretty close to the theoretical graph.
The case is that any feedback closed over non-linear circuit creates new harmonics, not present at output signal without feedback. And high order (new ones) are less reduced by feedback action because of lower loopgain there. That's probably what is JC pointing at.
P.S.: Blowtorch is a good example of the circuit that is quite free from high order harmonics.
P.S.: Blowtorch is a good example of the circuit that is quite free from high order harmonics.
On the other hand, here:
http://stereophile.com/solidpoweramps/407ayre/index4.html
one can also see high order harmonic content (more than at Creek amp). Frankly speaking, where is the proof of "no high order content" of certain designs?
http://stereophile.com/solidpoweramps/407ayre/index4.html
one can also see high order harmonic content (more than at Creek amp). Frankly speaking, where is the proof of "no high order content" of certain designs?
Hi all
Baxandall was right... it is well worth (re-) reading his articles.
His calculations were for a FET. The distortion gets even worse for a bjt.
This is why it is important to have high nfb - because too little makes things worse (unless you have managed to linearise the amp first).
Many 1970's amps boasting 0.1% distortion (!?) actually demonstrated this non-linearity when you measured their IMD. TUrns out many were around 0.4%. No wonder people didn't like 1970's amps much. The IMD is a better indication of non-linearity...
cheers
John
Baxandall was right... it is well worth (re-) reading his articles.
His calculations were for a FET. The distortion gets even worse for a bjt.
This is why it is important to have high nfb - because too little makes things worse (unless you have managed to linearise the amp first).
Many 1970's amps boasting 0.1% distortion (!?) actually demonstrated this non-linearity when you measured their IMD. TUrns out many were around 0.4%. No wonder people didn't like 1970's amps much. The IMD is a better indication of non-linearity...
cheers
John
(Yes, the graphs for the Creek and the Ayre were made at different power levels, but...)
It's obvious that the harmonic content of the two amps are wildly different. Compare the bottom trace of Fig. 6 for the Creek with the bottom traces in either Fig. 6 or Fig 7 for the Ayre. The harmonics left after notching out the fundamental approach a sawtooth on the Creek. Not good.
This is not mysterious. It's Fourier. One of the few things they taught me in school that turned out to be worth a damn when applied to real world audio electronics. (Egad, the stuff I had to unlearn once I started listening instead of reading spec sheets...)
The idea that "obviously" the answer is even more feedback is due to a flawed interpretation of the NFB/harmonic distortion % graph given above. Note that in a decent design the higher harmonics don't even exist until the feedback creates them. Surely "no" higher harmonics beats "low" higher harmonics any day of the week...not to mention the loss of imaging, etc. that results from high levels of negative feedback.
Charles,
I have adopted the rather lonely position of wide bandwidth (which I define as roughly 200-250kHz minimum) and low-to-no NFB and been raked over the coals over it numerous times. I'm glad to see someone with more "street cred" espouse the same ideals. To me, it was obvious once I started listening.
Yes, Nelson and John use little to no feedback when they can, but it seems to drive your entire philosophy, which is something I find interesting.
Grey
It's obvious that the harmonic content of the two amps are wildly different. Compare the bottom trace of Fig. 6 for the Creek with the bottom traces in either Fig. 6 or Fig 7 for the Ayre. The harmonics left after notching out the fundamental approach a sawtooth on the Creek. Not good.
This is not mysterious. It's Fourier. One of the few things they taught me in school that turned out to be worth a damn when applied to real world audio electronics. (Egad, the stuff I had to unlearn once I started listening instead of reading spec sheets...)
The idea that "obviously" the answer is even more feedback is due to a flawed interpretation of the NFB/harmonic distortion % graph given above. Note that in a decent design the higher harmonics don't even exist until the feedback creates them. Surely "no" higher harmonics beats "low" higher harmonics any day of the week...not to mention the loss of imaging, etc. that results from high levels of negative feedback.
Charles,
I have adopted the rather lonely position of wide bandwidth (which I define as roughly 200-250kHz minimum) and low-to-no NFB and been raked over the coals over it numerous times. I'm glad to see someone with more "street cred" espouse the same ideals. To me, it was obvious once I started listening.
Yes, Nelson and John use little to no feedback when they can, but it seems to drive your entire philosophy, which is something I find interesting.
Grey
estuart said:
Thanks Edmond,
That is interesting, but how do you get the values of the track inductances? I thought even if I could calculate empirically or measure (for which I do not have the instrumentation), there is also some unknown mutual inductance, capacitance etc. So I chickened out and tweak - if one can call it that - which is time-consuming and sometimes dangerous. (The wrong move and 1usec later your power devices depart.)
Any practical ideas from your side, please?
Fellow designers, for some reason, many of you cannot understand WHY we try to use as little feedback as possible, rather than as much feedback as possible. Could you possibly believe after 30-40 years of design that we have NOT tried high feedback? Did we not also attend college and learn the wonderful effects and the advantages of negative feedback?
What happened is that we were disappointed with the results of our high feedback amplifiers. That's why we cut back, ideally to no global feedback. However, to do this, we must keep our higher harmonics low and our overall distortion OK. We also have to fight off power supply noise, thermal modulation, high output impedance, and who knows what else! It is a worthwhile challenge for us, when people hear and love our efforts.
What happened is that we were disappointed with the results of our high feedback amplifiers. That's why we cut back, ideally to no global feedback. However, to do this, we must keep our higher harmonics low and our overall distortion OK. We also have to fight off power supply noise, thermal modulation, high output impedance, and who knows what else! It is a worthwhile challenge for us, when people hear and love our efforts.
john curl said:
John,
I would happily share your sentiment about the feedback, however you probably would also agree that a non-feedback amplifier is much more expensive to make in production and so everything comes down to price after all. I too did experiment with no overall NFB and "no feedback at all" amplifiers but for a budget mass produced amplifer this approach is very difficult if not impossible to implement so that the advantages of having no feedback would not be overrun by the disadvantages of a cheap implementation. By the way, Creek 5350SE was a recommended in "class A" component in Stereophile and a Component of the Year 2001 - if I remember correctly.
Alex
Hi
As a modest "hobbiest"
I must agree about NFB. It does not really "linearize" the amp. It cannot 'correct' for such higher order harmonics and actually seems to create them. IMHO having linearity with each stage, as much as possible anyway, is more important. This thread is geared towards output devices and topology, but what are your opinions of employing correction to all stages, drastically risking complexity, but using diffent types of error feedback or feedforward to each stage. Such as Dr Hawksford's paper on error correction for small signal stages, differentials, ect. Generally, improving linearity of each stage sould improve distortion figures, by reducing the non-linear components with-in the NFBL before the output stage. Or is this type of fix worth too little for too much
??
As a modest "hobbiest"
I must agree about NFB. It does not really "linearize" the amp. It cannot 'correct' for such higher order harmonics and actually seems to create them. IMHO having linearity with each stage, as much as possible anyway, is more important. This thread is geared towards output devices and topology, but what are your opinions of employing correction to all stages, drastically risking complexity, but using diffent types of error feedback or feedforward to each stage. Such as Dr Hawksford's paper on error correction for small signal stages, differentials, ect. Generally, improving linearity of each stage sould improve distortion figures, by reducing the non-linear components with-in the NFBL before the output stage. Or is this type of fix worth too little for too much ??GRollins said:
Well, I tend to be a "black and white" kind of guy. If feedback is good, you should use as much as possible. And if it's not good, you shouldn't use any.
I think the idea that feedback is like salt in your food, and that there is an ideal amount (the "Goldilocks and the three bears" theory) is just plain silly.
Nelson Pass said:
One time I was talking to Ken Stevens (Convergent Audio Technology), and he said the more feedback the better, as long as you didn't do something stupid to achieve it (like adding an extra gain stage). I don't agree with his philosophy, but I have to give him credit for at least having an intellectual consistency.
If there's one thing I don't have patience for, it is intellectual inconsistency. Like the speakers with diamond tweeters and flaccid cones. Or the zero feedback preamps paired with "low feedback" power amps.