Let's suppose we build a 50 Watt power amplifier that has 20 mV input sensitivity in open loop, with an open loop gain of 60 dB. Then we apply the proper amount of GNF to set the input sensitivity to 1V and reduce distortion.
The same input sensitivity can be achived by applying local negative feedback in each gain stages. The distortion of each stage will be lower, too. The end result will be the same input sensitivity as in the GNF case, and probably the distortion will be in the same magnitude of order.
I think the harmonic spectrum of the distortion will be different.
In the GNF case each stage gets badly distorted signal from the previous stage, and only using a huge amount of GNF can we manage to have a clean signal at the output. In contrast, in the local feedback case each stage will send a low-distortion signal to the next stage.
What is your opinion? Do we destroy the noise performance with local feedback? Do we have better gain distribution between stages? Which feedback will give better sound? (I think local feedback wins, but I don't have any hard evidence).
The same input sensitivity can be achived by applying local negative feedback in each gain stages. The distortion of each stage will be lower, too. The end result will be the same input sensitivity as in the GNF case, and probably the distortion will be in the same magnitude of order.
I think the harmonic spectrum of the distortion will be different.
In the GNF case each stage gets badly distorted signal from the previous stage, and only using a huge amount of GNF can we manage to have a clean signal at the output. In contrast, in the local feedback case each stage will send a low-distortion signal to the next stage.
What is your opinion? Do we destroy the noise performance with local feedback? Do we have better gain distribution between stages? Which feedback will give better sound? (I think local feedback wins, but I don't have any hard evidence).
IMHO: It depends. How many stages between input and output? A combination of both may well be the answer.
A two stage single ended amplifier may behave better with only local feedback.
Tests have shown our ears distort a considerable amount but our ears are very sensitive to phase shifts.
The more stages there are the more phase shift exits. A side effect is that more phase shift will increase the possibility of instability with increasing amounts of GNFB.
A two stage single ended amplifier may behave better with only local feedback.
Tests have shown our ears distort a considerable amount but our ears are very sensitive to phase shifts.
The more stages there are the more phase shift exits. A side effect is that more phase shift will increase the possibility of instability with increasing amounts of GNFB.
I've built very few Hi-Fi amps and done a bit more simulations.
Most probably everyone will give you some personal preference based on something he has read or tried or both.
Some here have written too, and they are a very huge source of information.
Personally I'd go for the lowest possible number of stages (usually this means two in tube world) and apply some local feedback.
I'd then apply (switchable) differential global feedback.
Most probably everyone will give you some personal preference based on something he has read or tried or both.
Some here have written too, and they are a very huge source of information.
Personally I'd go for the lowest possible number of stages (usually this means two in tube world) and apply some local feedback.
I'd then apply (switchable) differential global feedback.
Control systems design theory shows that the two cases described are absolutely identical as far as final distortion is concerned.
The equivalent block schematics can be changed to one or the other with the outcome being the same.
But in practice it does depend on the gain and overload characteristics of each stage. In the two cases, the input signals to each stage will be different and that will modify the distortion characteristics.
That also means you can't give a general answer to this question, it all depends on the practical realizations of the circuitry.
Jan
The equivalent block schematics can be changed to one or the other with the outcome being the same.
But in practice it does depend on the gain and overload characteristics of each stage. In the two cases, the input signals to each stage will be different and that will modify the distortion characteristics.
That also means you can't give a general answer to this question, it all depends on the practical realizations of the circuitry.
Jan
With global feedback, all stages except the last only have to produce far smaller signals than with local feedback, and therefore distort less.
With due respect: I would argue with that. As the saying goes: It all depends.
I'm able to achieve better results with less amplification and local feedback than with global negative feedback depending on output tube and topology used. YMMV
Negative feedback applied in parallel with signal doesn't increase voltage demands, but does increase power (current) demands. Negative feedback applied in series with signal must by definition increase voltage demands.
All good fortune,
Chris
All good fortune,
Chris
Speaking in generalities is meaningless. It depends on the amplifier topology. For a typical VFA, the stages before the compensation have to have local degen so that they don't clip on fast transients. With global feedback, the signal level at each stage depends on the frequency. Amplifier topology dictates where degen/local feedback can or cannot be put. You do not want to put too much in the output stage where it will compromise efficiency, or in the VAS where it wastes voltage swing, although large emitter resistors do help thermal stability. Many amplifiers are unstable because the global feedback has been pushed too far in pursuit of low THD. Such amps are very fussy about the load. This is less likely with local degeneration, but not impossible.
Analogies from modern semicon amplifier design aren't necessarily useful for valve amplifiers. A conventional modern semicon amp has a transconductance first stage feeding a heavily compensated (degenerated by collector-to-base capacitor) VAS that provides essentially all of the voltage gain, and a hopefully flat-ish double or triple output voltage follower.
By contrast, a conventional modern valve amp has several stages, each of modest gain and with comparatively minimal compensation, and each providing some voltage gain. Degeneration at any stage means the preceding stage(s) must provide that same amount of extra drive. No escaping the Second Law.
All good fortune,
Chris
By contrast, a conventional modern valve amp has several stages, each of modest gain and with comparatively minimal compensation, and each providing some voltage gain. Degeneration at any stage means the preceding stage(s) must provide that same amount of extra drive. No escaping the Second Law.
All good fortune,
Chris
Ernst Nordholt wrote some interesting things about it in his PhD thesis. If I remember well, his line of reasoning was something like this:
-You can reduce the distortion of a stage by local or by global feedback. When you use local feedback, you make it more difficult for the preceding stages, because they then have to handle larger signals. Hence, global feedback is in principle preferable.
-How much global feedback you can apply is limited by stability issues. If you have to reduce global feedback for stability reasons, it is better to do that by local feedback than by bluntly throwing away overall loop gain without getting any local feedback in return.
For example, in the conventional transistor amplifiers that Steve and Chris wrote about, you at least have local feedback in the output stage (an emitter follower is a stage with 100 % local feedback, hence its low voltage gain) and via the Miller capacitor in the preceding stage. The local feedback in the output stage is the reason why the preceding stage has to handle large voltage swings, it wouldn't if the output stage had been common emitter. The Miller compensation reduces overall loop gain in the frequency range where it needs to be reduced, and does so by local feedback. It's unfortunate that the output stage's local feedback is active at all frequencies.
-You can reduce the distortion of a stage by local or by global feedback. When you use local feedback, you make it more difficult for the preceding stages, because they then have to handle larger signals. Hence, global feedback is in principle preferable.
-How much global feedback you can apply is limited by stability issues. If you have to reduce global feedback for stability reasons, it is better to do that by local feedback than by bluntly throwing away overall loop gain without getting any local feedback in return.
For example, in the conventional transistor amplifiers that Steve and Chris wrote about, you at least have local feedback in the output stage (an emitter follower is a stage with 100 % local feedback, hence its low voltage gain) and via the Miller capacitor in the preceding stage. The local feedback in the output stage is the reason why the preceding stage has to handle large voltage swings, it wouldn't if the output stage had been common emitter. The Miller compensation reduces overall loop gain in the frequency range where it needs to be reduced, and does so by local feedback. It's unfortunate that the output stage's local feedback is active at all frequencies.
There have been semicon amps made with common emitter output stages, gain set to maybe 2. If I understand it correctly, this complicates crossover issues (not sure why, though) but does allow the VAS to provide full drive with V+ and V- shared (decoupled) from the output supplies. That can't quite be done with voltage follower outputs, so some cost savings. But I'm far afield now.
All good fortune,
Chris
All good fortune,
Chris
The emitter resistors of a complementary emitter follower stage cause a distortion compensation effect in the crossover region when you bias the transistors at some optimal current. As far as I know, that doesn't work that well with a CFB stage. In any case, CFB stages also have local feedback.
A couple of weeks ago, someone posted a valve line-level transformer driver circuit with a very tiny amount of local feedback around the output stage. It turned out that the reason for that was to increase the distortion of the stage in front of it, because that distortion was in antiphase with the output stage's distortion and they largely cancelled when you used the right amount of local feedback.
Subtle distortion compensations like this are not covered in Nordholt's thesis.
A couple of weeks ago, someone posted a valve line-level transformer driver circuit with a very tiny amount of local feedback around the output stage. It turned out that the reason for that was to increase the distortion of the stage in front of it, because that distortion was in antiphase with the output stage's distortion and they largely cancelled when you used the right amount of local feedback.
Subtle distortion compensations like this are not covered in Nordholt's thesis.
Let's think about typical, simple two stage SE amplifier. At full power the voltage amplifying stage generates some 0.5 % THD while the output stage generates 5%. Then we apply 12 dB (4-times) GNFB. The result is ~1.25% THD at full output power.
Second option; we apply 6 dB local NFB in both stages (12 dB total). The result is 2.5 % THD at full output power.
Which one is better...?
Second option; we apply 6 dB local NFB in both stages (12 dB total). The result is 2.5 % THD at full output power.
Which one is better...?
Modelling shows: (but even the best plans will change during implementation - theory and practise are often not the same although they can be pretty close. Modelling is still better than using a slide rule.)
Scenario 1: Output stage 7868 with 43% UL. Driver stage: SRPP E81CC, amplification approx. 30. GNFB from secondary.
Scenario 2: Output stage 7838 with 43% UL and 8% cathode feedback. Identical SRPP driver setup, no GNFB.
for Chris:
In order to obtain similar distortion with scenario 1 more amplification in the driver stage is required to offset the GNFB, even the E81CC in SRPP is not sufficient.
for Jan.Didden:
Scenario 2 has less than half the distortion of scenario 1 (at mximum output) and has fractional more output at the same level of input from source. (less distortion for both: at 1 Watt, and also at onset of grid grid at maximum output) Modelling shows 0.31% scenario 1 and 0.25% scenario 2 at 1 Watt and 1.09% scenario 1 and 0.54% scenario 2 at onset grid current.
The unknown part in this is the impact of the output transformer in both scenario's but I'll find out in due time. Still waiting on some parts (some took 9 months to arrive due to shipping issues and minimized air traffic), had hopes on completion by Easter but looks to be the middle of the year by now.
YMMV
Last edited:
I often say that every discussion of feedback will devolve into a lower information state, as demanded by entropy. EL506 has given an excellent example of the difficulty keeping apples and oranges separated, the output transformer (OPT). Special case distortion components cancelling each other is another (but let's ignore that).
smoking-amp has lately been campaigning for a slightly negative source impedance driving the OPT - just enough to cancel bulk resistance of the windings. Even if not taken all the way to negative, lowering the driving impedance lowers OPT distortion and improves bandwidth, so is its own Good Thing. We should include that as an ingredient of the mix.
A combination of voltage sensing negative feedback and current sensing positive feedback taken from the output valve / OPT junction can get OPT driving impedance down to any stable number desired. If both are brought back to an earlier, presumably lower distortion, stage, the extra open loop gain from the positive feedback will increase the amount of negative feedback available from a given topo. Given that most open loop gain is from earlier, presumably lower distortion stage(s) that's a net lower closed loop distortion.
All good fortune,
Chris
smoking-amp has lately been campaigning for a slightly negative source impedance driving the OPT - just enough to cancel bulk resistance of the windings. Even if not taken all the way to negative, lowering the driving impedance lowers OPT distortion and improves bandwidth, so is its own Good Thing. We should include that as an ingredient of the mix.
A combination of voltage sensing negative feedback and current sensing positive feedback taken from the output valve / OPT junction can get OPT driving impedance down to any stable number desired. If both are brought back to an earlier, presumably lower distortion, stage, the extra open loop gain from the positive feedback will increase the amount of negative feedback available from a given topo. Given that most open loop gain is from earlier, presumably lower distortion stage(s) that's a net lower closed loop distortion.
All good fortune,
Chris
Thanks for the useful inputs. Let's simplify things to a 3-stage voltage amplifier with line-level output. The voltage gain of each stage is 10. In the A version let's degenerate each stage by 6 dB local negative feedback so the the gain will be 5 for each stage, 5 x 5 x 5 = 125 globally. In the B case let's apply 18 dB global negative feedback. Do we get the same distortion at the same output level? And more importantly: do we get the same FFT harmonics spectrum in the two cases? Somewhere I read that GNFB reduces the net harmonic distortion, but the higher order spectral component will increase over 2nd harmonic. So there will be a spectral imbalance that is unpleasant to human hearing. Is it true?
Instead of answers, I have one question:
What has been the trend of feedback purpose in different era of Hi-Fi, and what are current trends?
Lower as much as possible THD, chase specific harmonic slope, chase specific predominance of harmonics and phase, chase pleasant ovedrive on peaks, maximize power and damping factor, others?
What is each of us chasing?
Thanks
What has been the trend of feedback purpose in different era of Hi-Fi, and what are current trends?
Lower as much as possible THD, chase specific harmonic slope, chase specific predominance of harmonics and phase, chase pleasant ovedrive on peaks, maximize power and damping factor, others?
What is each of us chasing?
Thanks
In tubes, distortion is very much proportional to signal level. This means nearly all the distortion is produced in the output stage. Hence local feedback will reduce this much less than global feedback will. Hence global feedback will reduce the distortion much more than local feedback around each stage.
Cheers
Ian
Cheers
Ian