It's like everything else in the real world: a trade-off. I did a project based on 807s. This type has a spectacularly low THD figure in the spec sheet (1.8%). However, when doing the Twin-T test, the residual is highly distorted, indicating a lot of high order harmonics. Listening open loop bears this out, as there is quite a bit of that pentode nastiness. With some program material, it's as irritating as fingernails on a black board. The developer of the type, O. Schade, recommended feeding back 10% of the plate voltage in a local feedback loop. This advice is definitely spot-on, and goes a long way to fixing the problem. All that was required was about 6.0db(v) of gNFB to get the woofers under control and to take off any remaining edginess.
What you have to give up is sound stage. That's where all that near noise floor harmonic mess that Crowhurst mentions comes in. You lose the 3-D effect that is quite apparent while running open loop. Tighter bass and no more dissonant nastiness is a worthy trade.
Of course, you can overdo the gNFB, and the result is a definite solid statey sound: lifeless, monotonic bass, loss of fine background detail, lyrics become noticeably more difficult to understand. 20db(v) of gNFB was hideous, and in some ways, even worse than no NFB at all with the 807 project.
For another project that uses 6BQ6s, you don't have so much of that pentode nastiness since this type mainly makes h3. I included variable gNFB (0 - 13db(v)) but no local NFB. Adding the full 13db of gNFB is definitely solid statey, and takes the life out of Techno and Metal. About 6.0db(v) sounds about right. Less sounds edgy, but with more sound stage.
Gain something here; lose something there. Is what you gain worth more than what you lose? What's worse is trying to cover up a bad open loop design with excessive gNFB.
You have to decide whether what you have to give up makes what you gain worth it? For pentode finals, I definitely think it is.
Actually, the articles which demonstrate the way NFB multiplies nth-order distortions are good science provided they are correctly understood and taken in context. I remember reading such an article in Wireless World many years ago (by Baxandall? I can't remember). Three things follow from this:
1. You can't make a bad amp good by slapping on lots of NFB (global or local - the maths work for both!).
2. For a good amp with moderate levels of distortion, sufficient NFB can reduce low-order distortion while not increasing high-order distortion to more than it already was due to inherent device nonlinearity.
3. Small amounts of NFB can make things worse, by adding more high-order products but not reducing low-order by much. This means that it may be daft to add 6dB of feedback to a very good amp, but it may be wise to add 20dB of feedback to a moderately good amp.
These theoretical findings do seem to accord with what people hear. High feedback amps often sound horrible. Zero-feedback amps can sound good, but a little feedback makes them sound worse - unfortunately people then reach the wrong conclusion and believe that if a little feedback is bad then more feedback must be worse. The best level of feedback is the right level for that application of that circuit.
1. You can't make a bad amp good by slapping on lots of NFB (global or local - the maths work for both!).
2. For a good amp with moderate levels of distortion, sufficient NFB can reduce low-order distortion while not increasing high-order distortion to more than it already was due to inherent device nonlinearity.
3. Small amounts of NFB can make things worse, by adding more high-order products but not reducing low-order by much. This means that it may be daft to add 6dB of feedback to a very good amp, but it may be wise to add 20dB of feedback to a moderately good amp.
These theoretical findings do seem to accord with what people hear. High feedback amps often sound horrible. Zero-feedback amps can sound good, but a little feedback makes them sound worse - unfortunately people then reach the wrong conclusion and believe that if a little feedback is bad then more feedback must be worse. The best level of feedback is the right level for that application of that circuit.
Member
Joined 2009
Paid Member
This is what I've read and understood too and I employed this approach with my last SS amp with good results.
Nowadays, I'm aiming for zero gnfb. It's a fun/interesting design challenge. It makes you realize very quickly why gnf is so popular.
What kind of circuit is that graph from?
If OLG continually decreases, or slants downward as is the case with super-high OLG opamps, lower order harmonics will be corrected more than high order ones. Thus low order harmonics will lose magnitude compared to higher order ones.
- keantoken
If OLG continually decreases, or slants downward as is the case with super-high OLG opamps, lower order harmonics will be corrected more than high order ones. Thus low order harmonics will lose magnitude compared to higher order ones.
- keantoken
This graph is from an ideal circuit from some rose reality that has no other artifacts than 10% of 2'nd order harmonic. The real life is much dirtier, of course!
The Royal Device and Martin Collums articles are the ones that are particularly unimpressive. They contain no math and loads of conjecture.
Crowhurst and Baxandall on the other hand provide rigorous treatment. The shift in harmonic structure is indeed pure math; it's well known, but I don't remember anything about time delay as a distortion mechanism other than phase margin and stability. I guess I need to read them again.
I do hope you're not lumping the Collums and Royal Device articles in with Baxandall and Crowhurst. I am certainly not.
My vote (really an observation) is Crowhurst and Baxandall contain math, Collums and Royal Device contain pseudoscience and technobabble.
Cheers,
Michael
BTW, I personally have not yet built an amp that needs global negative feedback. My own design approach is to use combination of device operating region selection and local feedback with simple topologies to realize simple harmonic structure and good speaker damping. I have discovered that many distortion cancellation mechanisms are also distortion conversion mechanisms which shift some energy to higher harmonics. I do agree that lower harmonics are in general more natural sounding. Air and ears create large amounts of low order distortion after all (he says whilst waving his hands...) seriously I don't study much psychoacoustics but I know what I like ;-)
Last edited:
I don't see how a circuit producing only even harmonics can produce odd harmonics when given feedback.
- keantoken
- keantoken
The feedback contains both fundamental and 2nd harmonic. The second-order distortion in the amplifier forward path then creates some 3rd by combining 1st and 2nd, etc, etc, etc. You get an infinite number of products!
For those into maths (a minority, I know!) this is related to the infinite series:
1/(1+x) = 1 -x +x^2 -x^3 +x^4 . . .
Something like this appears if you put an amplifier with some second-order distortion in the normal formula for gain with feedback.
As Wavebourn's graph shows, if you are going to add any feedback you need to add enough to depress the products you have just created. In reality its not quite as bad as that, as the devices will already have some high-order products anyway. How come we can make any amp sound reasonable? Two things on our side:
1. for small signals the inherent distortion is much less than 10%, so the extra distortion is much less too and it gets better with higher order. If you can halve your 2nd-order by better design, then the 3rd will be down to a quarter etc.
2. triodes have unusual feedback, because in an ideal triode the feedback has the same distortion as the forward path so it exactly cancels. Unfortunately you can't buy an ideal triode but some are quite good when used with a high impedance load.
For those into maths (a minority, I know!) this is related to the infinite series:
1/(1+x) = 1 -x +x^2 -x^3 +x^4 . . .
Something like this appears if you put an amplifier with some second-order distortion in the normal formula for gain with feedback.
As Wavebourn's graph shows, if you are going to add any feedback you need to add enough to depress the products you have just created. In reality its not quite as bad as that, as the devices will already have some high-order products anyway. How come we can make any amp sound reasonable? Two things on our side:
1. for small signals the inherent distortion is much less than 10%, so the extra distortion is much less too and it gets better with higher order. If you can halve your 2nd-order by better design, then the 3rd will be down to a quarter etc.
2. triodes have unusual feedback, because in an ideal triode the feedback has the same distortion as the forward path so it exactly cancels. Unfortunately you can't buy an ideal triode but some are quite good when used with a high impedance load.
The infinite series is perhaps what some people mean by "the feedback goes round and round" ...woh-o-oh -oooh-oh, and it comes out here 😉 But it's not caused by delay, so the round and round metaphor is IMO at least misleading.
One can approximate an ideal triode by connecting a pentode with local plate-grid feedback. This can be used with a relatively low impedance load due to the effective plate resistance approaching 1/gm.
Thanks!
Michael
Actually, both. When only one stage is surrounded by feedback you may pay less attentions on phase shifts.
Member
Joined 2009
Paid Member
gnf is like taking antibiotics - take too few and you create a problem, you have to take enough to kill the bugs or best don't take any (same goes for one of my friends who uses vodka in place of antibiotics!)
Tell that to someone I know...wasn't happy until he reached 50 dB of feedback. 30 dB gNFB and 20dB local.
Let's be quite clear about this: Is it delayed feedback which causes the infinite series of harmonics to be generated? Or is it something else?
Feedback causes them linearizing transfer curves. Phase shifts in feedback loop cause uneven sidebands of intermodulation (really nasty, lifeless, dry sound, like sand in the mouth!).
Yes, but my question is very specific:
Is the effect shown in these curves caused by delay in the feedback?
Also, do you have the mathematical model from which these curves are generated?
Attachments
I guess by people you mean me, as I'm people 😉 I meant that it goes round and round.
It must go round and round because it's fed back, in a loop. The output is fed back (hence the term feedback!) into the input. If you can detect that feedback is in use then it must be changing the output, which in turn gets fed back to the input. From there it works its way through that amp to the output and gets fed back to the input. And round and round it goes.
OK you say, but each time around the feedback artifacts get much quieter so it doesn't travel round and round for long. Well yes and no, it will travel round until you hit the noise floor deep enough to be dissipated. How audible it is on say the 3rd time around will depend on factors of course.
In theory I guess it doesn't go round and round. But in that theory I think you also have a linear amplifier with zero time delay. As soon as you look at a real non-linear amp with a time delay between the input and the output that theory becomes junk.
The non-linearities will give you the multiplication of harmonics, the time delay will give you phase shifts that vary with frequency. Negative feedback is all about correcting the input signal by comparing it to the output signal. When the output signal is for instance travelling through the amp you cannot compare it properly because it is delayed in time.
Have a read of the Audio Note page here which has an interesting take on the overload effect of negative feedback due to the time delay it takes the signal to transit the amplifier.
- Status
- Not open for further replies.