It's a Baxandall graph, not Linsley-Hood as SY noted.
The graph is a result of feeding back a signal through a square-law device. Any harmonics in the feedback get again distorted by the square law, go through the feedback again, get distorted etc etc. So initially, when feeding back, you get a whole slew of additional harmonics before the feedback gets so powerfull that they all get supressed.
In your case, if you don't have the square law, if your amp is already reasonably linear, the feedback products don't get distorted again so much because your amp is reasonably linear.
The original Baxandall graph was for a single non-degenerated FET device which, as we know, has an almost perfectly square-law transfer curve.
BTW a pure square-law would be like Vout = A*Vin^2. After that, it's just mathematics
jan didden
The graph is a result of feeding back a signal through a square-law device. Any harmonics in the feedback get again distorted by the square law, go through the feedback again, get distorted etc etc. So initially, when feeding back, you get a whole slew of additional harmonics before the feedback gets so powerfull that they all get supressed.
In your case, if you don't have the square law, if your amp is already reasonably linear, the feedback products don't get distorted again so much because your amp is reasonably linear.
The original Baxandall graph was for a single non-degenerated FET device which, as we know, has an almost perfectly square-law transfer curve.
BTW a pure square-law would be like Vout = A*Vin^2. After that, it's just mathematics
jan didden
Last edited:
You are right (but remember: you have a phase splitter with a lot of local feedback). However it was not my intension to make this an academic discussion. I also find your experiment interesting.
I can not see how the high local feedback at the phase splitter could have effect to whole amplifier.
My explanation to the results I got is that tubes (I used and how those are biased) form a very linear amplifier chain and therefor differs very much from square-law components like fets and transistors.
The phase splitter is one of the stages that together determine the transfer curve of the amp. If you make one very linear, the whole amp gets a bit more linear.
jan didden
Global Negative feedback looks great on paper, with a 1kHZ test signal. In the real world there are many variables; topology, device choice, local or degenerative feedbacks, slewrates and miller effects of each device, etc. The increased higher order harmonics that I saw with increased GFB may be due to the limitations of the feedback loop in trying to deal with supersonic frequency energies generated by the non perfect front end differential amplifier (or equiv.) (I.M. products included), or the rolloff caused by the output devices, thereby undoing the corrective efforts at those higher frequencies. It may be a combination of several things. You never want to ask a feedback loop to correct something it can't correct, for whatever reason, or it may well create some new distortions; spurious oscillation, I.M. products, slewing, overdrive/clipping of internal stages, or excessive smoke and funny popping sounds. Oh, was that your $5000 speaker? Sorry.
Well I do agree with you but you make it sound as if it is some kind of black magic. It isn't. It is all fully understood for many decades and competent designers know how to design a feedback amp to stay clear of the problems that could pop up if you don't follow the well-known rules.
jan didden
jan didden
It may be true that some amplifiers (a hypothetical square law transfer function) will have higher harmonic distortions rise before they ultimately fall, but it seems that distortion reduction for the real world example of your amplifier can be equally beneficial for all harmonics and all amounts. Your curve shows every harmonic falling more or less by the same amount as the feedback factor.
Very interesting result.
Those that want to proclaim that distortion is unbeneficial in small amounts need to be very specific about the conditions they are talking about. It ain't generally the case.
By the way, the presence of local feedback altering the comparison is a red herring. Each active device can be considered a black box with a transfer function. Within that black box might be a given device with its own feedback altering its initial characteristics. That wouldn't matter, the comparison is still valid. Transfer curves of pentodes with local feedback are one interesting example, where they start to look more like triodes.
David S.
Let me respectfully disagree. My nice sounding Swinik (class A+C amp) would sound extremely horribly without all that multiple NFB loops.
It is fairly low in absolute level and 20 dB below the dominant 2nd harmonic.
I once did a curve fit to a spider (woofer suspension) force vs. displacement curve. The curve was very smooth, esentially a straight central area with nicely rounded progressive stiffening at the ends. The curve fit didn't come even close to looking like the spider transfer function until the fitted curve included 7th order terms.
Every harmonic analysis that I've seen of a single ended amp at moderate output has a pretty full series of harmonics, even though 2nd will be strongest. The rest are there at a diminishing level.
David S.
I once did a curve fit to a spider (woofer suspension) force vs. displacement curve. The curve was very smooth, esentially a straight central area with nicely rounded progressive stiffening at the ends. The curve fit didn't come even close to looking like the spider transfer function until the fitted curve included 7th order terms.
Every harmonic analysis that I've seen of a single ended amp at moderate output has a pretty full series of harmonics, even though 2nd will be strongest. The rest are there at a diminishing level.
David S.
If it has 2 or more stages their transfer functions will be multiplied, right? A* A^6 = A^ 7.
Ok, if you don't like it a^3 * a^4 = a^7
Right?
Also, tubes have geometry. Control grids are not virtual, they made of real wire, so one tube can be viewed as many tubes in parallel, and the higher is negative voltage on the grid, the more the tube looks like many tubes with different parameters connected in parallel. It is one of sources of nasty distortions in tubes. If to make grids of thinner wire, more dense, and further from cathodes, more of voltage and (surprise!) current is needed to control them, but as the result their transfer function is smoother. Examples are: 300B, 2A3, 4P1L, GM-70
Depends on the tubes and the topology. That's why it's easy to design an acceptable tube amp but a lot harder to design an excellent one.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.