Reread the article. Only pure square-law components(*) show marked increase of high harmonics with small amounts of feedback(+), other non-linearities benefit immediately from feedback as they have intermodulated higher-order harmonics that respond much more rapidly to linearization.
All devices show total harmonics falling with feedback.
All devices have improved input and output impedances and flatter frequency response with feedback too, amongst the other advantages.
And of course intermodulation products are not harmonic, and seldom mentioned by people who think some harmonics are "good".
(*) which don't exist, but FETs are closest, not valves.
(+) and only because they don't have any 3rd/4th/5th to start with, and the dominant harmonic always falls.
I'm not talking about power output, of course the power delivered reduces at higher impedances. However most speakers are designed for nominally flat response (SPL) at constant voltage. If you have a peak in impedance driven by an amplifier with high output impedance, you will get a corresponding peak in voltage which will lead to a peak in SPL delivered. You can design speakers which give a nominally flat response when driven by high output impedance, however most are not designed this way.
The fact that many ( too many) speaker combinations is designed with constant voltage amps is a problem ! Two faulty procedures dont cancel !
Broadband speakers ( one element and no filter components) combined with current delivering amps shows what might be possible.
dunno where you get that from, a speaker is a linear motor.
Any motor you will try varies all kinds of parameters at different speeds, which is why modern tram and train motors are carefully controlled with big powerful SCRs to optimise performance especially at slow speeds demanding large current swings.
That's the definition in audio terms of DSPs, algorithms and current measurement/optimisation.
Whether you like it or not, that's how industrial motors (or speaker drivers) work.
😱
I am scratching my head when I read that.
If your output transformer reflects 6K ohm from 8ohms as A-A load.
If I suddenly change the load x 7.5 the value how come you don't get it that makes a A-A load of 45K.
Try swinging say 100V across first 6K then 45K, and convince me the power output will increase!
Physics didn't change simply because we would like it to. 🙄
If the speaker is at resonance, it could very well put out more SPL for less drive, no? And a speaker is a tuned device, we're not talking about a driver in free air. The box and crossover act like a "control" of sorts, no? It is tuned to the Q of the speaker, right?
Last edited:
The driving force of a speaker is dependent of the current flow. Not the voltage
applied across it![/QUOTE]
In which case the speaker needs to be driven by a pentode output stage which has +ve current FB. But we don't do that, we do the exact opposite.🙂
Several commercial amps have been built over time that incorporated a DF control. That varied +ve current FB & -ve voltage FB.
I tried building such an amp many years ago & used a short length of resistance wire to sample current to the loudspeaker for the -ve FB. Real interesting to watch the loudspeaker LF peak first disappear, then go the other way. Too much & the amp would oscillate.😱
applied across it![/QUOTE]
In which case the speaker needs to be driven by a pentode output stage which has +ve current FB. But we don't do that, we do the exact opposite.🙂
Several commercial amps have been built over time that incorporated a DF control. That varied +ve current FB & -ve voltage FB.
I tried building such an amp many years ago & used a short length of resistance wire to sample current to the loudspeaker for the -ve FB. Real interesting to watch the loudspeaker LF peak first disappear, then go the other way. Too much & the amp would oscillate.😱
This is all well known stuff, I'm not making it up myself. I clarified and said that there will be increased output from the speaker (i.e. SPL), not increased power output. I'll clarify again and say that this is at lower levels when you are not constrained by maximum primary voltage swing. If the amplifier output source impedance is 8 ohm (say from a pentode output stage with no NFB), and level is set such that you get say 1 V rms with infinite load impedance, you will get 0.5V rms into an 8 ohm load impedance and 0.88V rms into 60 ohms. So between the 8 ohm load and the 60 ohm load you will get a 4.9 dB increase in output voltage. In many speakers this will result in a corresponding 4.9 dB increase in SPL. This can sometimes be helpful to get a bit of extra bass at the box resonant frequency, but not when the peak is at the midrange due to crossover component interaction.
The generation of harmonics in an amplifier can be quite complex.
Just a few observations:
But please remember that "All Generalizations Have Exceptions".
Triodes roughly follow Childs law, at 2/3 power; not square (2 power) law.
Beam Power Tetrodes and Pentodes roughly follow a higher than 2/3 power law.
If one triode follows another (2 cascade stages), the 2nd harmonic of each stage is out of phase. That causes 3rd harmonic distortion. How much 3rd harmonic distortion is dependent on the % of the two stages.
If they are the same percentage, that tends to produce lots of the 3rd harmonic (that may not even been there in the beginning of either stage), and reduce the 2nd harmonic.
A push pull stage tends to reduce the 2nd harmonic, and increase the 3rd harmonic.
If a single stage has 5% 2nd harmonic distortion, and no 4th harmonic distortion, and then negative feedback is applied to cancel most of the 2nd harmonic, then the original 2nd harmonic error signal comes back to the input.
5% of 2nd harmonic that goes through the same stage again, tends to generate the 4th harmonic. You might get up to: 5% x 5% = 0.25% 4th harmonic.
If a tube amplifier with an output transformer is going to have 60dB, 80dB, or 100 dB of global negative feedback, we may end up with an oscillator, especially if the load is a reactive loudspeaker, and the load is not a resistor.
Saying that what we need is lots and lots and lots of feedback will likely cause the above condition.
There was a production solid state amplifier that did not even have an output transformer, and had lots and lots of negative feedback. It tested out to be wonderful on a resistive load. But all of the amplifiers that were sold came back, because reactive loudspeakers blew up the amp, the loudspeaker, or both.
Just a few observations:
But please remember that "All Generalizations Have Exceptions".
Triodes roughly follow Childs law, at 2/3 power; not square (2 power) law.
Beam Power Tetrodes and Pentodes roughly follow a higher than 2/3 power law.
If one triode follows another (2 cascade stages), the 2nd harmonic of each stage is out of phase. That causes 3rd harmonic distortion. How much 3rd harmonic distortion is dependent on the % of the two stages.
If they are the same percentage, that tends to produce lots of the 3rd harmonic (that may not even been there in the beginning of either stage), and reduce the 2nd harmonic.
A push pull stage tends to reduce the 2nd harmonic, and increase the 3rd harmonic.
If a single stage has 5% 2nd harmonic distortion, and no 4th harmonic distortion, and then negative feedback is applied to cancel most of the 2nd harmonic, then the original 2nd harmonic error signal comes back to the input.
5% of 2nd harmonic that goes through the same stage again, tends to generate the 4th harmonic. You might get up to: 5% x 5% = 0.25% 4th harmonic.
If a tube amplifier with an output transformer is going to have 60dB, 80dB, or 100 dB of global negative feedback, we may end up with an oscillator, especially if the load is a reactive loudspeaker, and the load is not a resistor.
Saying that what we need is lots and lots and lots of feedback will likely cause the above condition.
There was a production solid state amplifier that did not even have an output transformer, and had lots and lots of negative feedback. It tested out to be wonderful on a resistive load. But all of the amplifiers that were sold came back, because reactive loudspeakers blew up the amp, the loudspeaker, or both.
Last edited:
Dumping what is essentially a current source (a non fed-back pentode stage) into a loudspeaker will certainly produce more voltage at the higher impedance frequencies of the loudspeaker.
That is far different than a solid state amplifier that has a damping factor of 100 into the same loudspeaker.
That is far different than a solid state amplifier that has a damping factor of 100 into the same loudspeaker.
Given its brevity, only 30 posts, I'd be tempted to rate this as one of the silliest threads in quite a while.
Good fortune to all,
Chris
Good fortune to all,
Chris
THD falls with the addition of negative feedback.
That is because the negative feedback reduces the low order harmonics which happen to be the strongest, but those low order harmonic error terms are subjected to the same transfer curves, which generates upper order harmonics.
Crowhurst again.
Just another generalization . . .
That is because the negative feedback reduces the low order harmonics which happen to be the strongest, but those low order harmonic error terms are subjected to the same transfer curves, which generates upper order harmonics.
Crowhurst again.
Just another generalization . . .
It doesn't reduce anything at all.
Even harmonic distortion cancels out in the output transformer.
If you check in each anode circuit they are present in large amounts.
The levels of odd harmonic distortion there in the first place don't get cancelled out.
That mostly has to be tackled with running a valve in the most linear part of the curves, avoiding crossover distortion and using some NFB.
(which was the thinking behind DTN Williamson's design.)
Another technique,-
Ultralinear tapping modifies this by introducing negative feedback into the true anode circuit (the screen grid), while the virtual anode remains to collect the electron stream with a variable modified mu generated by a large voltage dependent gain component.
So, we swop a lot of the non linearity for dynamic compression....(inc compression of the distortion), and modification of the transfer curves in the class A region.
What's not to like?
A lot!
CFB operates by modifying the operating points of the valve by introducing a counter phase signal in the low impedance circuit.
This has a totally different effect from NFB in the high impedance anode circuit, it substantially alters the drive circuit requirements, in effect halving the gain of the OPV.
However all NFB results in a reduction in loop gain.
Even order harmonic cancellation again:-
You get exactly the same phenomenon in the paraphrase phase splitter, including cancellation of microphonic noise and hum, which is why using 2, not otherwise very linear 12AU7 for such an amplifier/phase splitter works very well.
Brimar shows this in its own design which has good gain, high bandwidth and low distortion.
The cascode circuit is yet another example of high gain & distortion inhibition inc the frequency non linear component (miller effect), reduction done by push-pull design.
many chays to skin a wicken.
Last edited:
"It doesn't reduce anything at all" is misleading.
Asymmetrical distortion (affecting positive and negative-going parts of the signal differently) to a sine wave causes even-order distortion components (2nd, 4th, 6th order etc).
Symmetrical distortion (affecting positive and negative-going parts of the signal equally) causes odd-order components (3rd, 5th, 7th etc).
A single ended stage produces almost purely even-order distortion due to the square-law-like function of a tube or FET (and exponential-like for BJTs).
If you consider a push-pull stage to just be two single-ended stages operated differentially, then when the two output signals are combined/summed by the transformer, odd-order distortion is produced instead as you have added an inverted version of an asymmetrically distorted signal to itself, therefore you get a symmetrically distorted signal.
Even order distortion will still however exist at a low level in practice, because the push and pull stages are never perfectly matched (or not perfectly combined by the transformer 🙂) so even-order components cannot be cancelled completely.
In a practical application the distortion should be overall lower for the push-pull stage versus single ended because each tube is doing half the work so can operate more linearly (current or voltage swing will be halved for the same overall voltage and/or power output). It also brings benefits with respect to cancelling out the DC bias current saturating the transformer core, lowering distortion contribution from the transformer itself.
Every improvement you can make to a single ended stage can also be performed to a push-pull stage. Ultra-linear operation, cacoding, NFB are just more techniques which can (and should) be used to improve an amplifier stage's distortion, regardless of it being SE or push-pull/differential.
Somewhat interestingly it is the same reason why adding global NFB produces odd-order distortion - because you're adding a inverted version of the asymmetrically-distorted signal to itself.
Asymmetrical distortion (affecting positive and negative-going parts of the signal differently) to a sine wave causes even-order distortion components (2nd, 4th, 6th order etc).
Symmetrical distortion (affecting positive and negative-going parts of the signal equally) causes odd-order components (3rd, 5th, 7th etc).
A single ended stage produces almost purely even-order distortion due to the square-law-like function of a tube or FET (and exponential-like for BJTs).
If you consider a push-pull stage to just be two single-ended stages operated differentially, then when the two output signals are combined/summed by the transformer, odd-order distortion is produced instead as you have added an inverted version of an asymmetrically distorted signal to itself, therefore you get a symmetrically distorted signal.
Even order distortion will still however exist at a low level in practice, because the push and pull stages are never perfectly matched (or not perfectly combined by the transformer 🙂) so even-order components cannot be cancelled completely.
In a practical application the distortion should be overall lower for the push-pull stage versus single ended because each tube is doing half the work so can operate more linearly (current or voltage swing will be halved for the same overall voltage and/or power output). It also brings benefits with respect to cancelling out the DC bias current saturating the transformer core, lowering distortion contribution from the transformer itself.
Every improvement you can make to a single ended stage can also be performed to a push-pull stage. Ultra-linear operation, cacoding, NFB are just more techniques which can (and should) be used to improve an amplifier stage's distortion, regardless of it being SE or push-pull/differential.
Somewhat interestingly it is the same reason why adding global NFB produces odd-order distortion - because you're adding a inverted version of the asymmetrically-distorted signal to itself.
Last edited:
who says?
Have you ever worked with PPP configurations and high quality mil spec valves?
I can assure you, with a little time and trouble you can get perfect matching, and the more valves are paralleled the better it gets.
There's lots of claims which are totally wrong.
It is constantly claimed the paraphrase phase splitter can't work well, because of inbalance in exactly the same way you try to describe, as well as claims over decades that valves always have dissimilar triodes in pairs such as the 6SN7/6SN7 etc.
Well I can assure you when I fed the 2 outputs of the phase invertor into my distortion meter and nulled out the distortion from the gain of the 2 halves what little distortion I could detect fell right down the scale to well below 0.1% at full output.
In fact the resistor divider network designed originally for correct balance was right in the ballpark, but I installed a pot to be sure.
By the same token I went ahead and swapped out valves several times (ya know, people call it "tube rolling") to see if I could detect any difference between those mystical round plates, square plates and all...
The HP distortion meter scarcely moved at all, there was about 0.02% THD difference, no matter what I put in there... right on the threshold of detection.
So I call BS on these kind of posts, as well as all those "tube rollers" with their myths and legends... 🙄
The biggest hassle with PP amps is crossover distortion with transconductance non linearity arriving at this point.
Working carefully to optimise it, usually results in far better linearity than any SE amp, but I usually found the 70% quiescent rule is wrong.
Best to watch it all on a 'scope, and the distortion meter, with careful attention to IMD, 'cos it's the most audible bad thing you will ever get... (if you're listening that is..)
Last edited:
Kirchoff's Laws
Most never think of Kirchoff when talking DF. The DC R of the loudspeaker is part of the output loop & sets an upper limit on what DF can do. So DF's greater than 20 don't mean much.
Its possible to build amps with -ve internal resistance to cancel external +ve resistance. But done by few, instability is always a possibility.😱
But a DF of 10 or 20 for sure is better than none.🙂
Most never think of Kirchoff when talking DF. The DC R of the loudspeaker is part of the output loop & sets an upper limit on what DF can do. So DF's greater than 20 don't mean much.
Its possible to build amps with -ve internal resistance to cancel external +ve resistance. But done by few, instability is always a possibility.😱
But a DF of 10 or 20 for sure is better than none.🙂
Perfect matching doesn't exist. You can only guarantee a 'perfectly matched' tube is as well matched as you are able to measurement it. Sure, you may luck out and they may be quite a bit better matched than you can measure, but you don't know if you don't have the equipment to measure. If you only need to match two devices, have the equipment capable of measuring to a tight tolerance (amplifier and THD meter counts as high precision equipment 😉) and a large enough pool of devices to test you can match to fairly high tolerance. If you're satisfied with 0.02% THD good for you 🙂. Cherry picking like that doesn't always work in a production environment though. My rule of thumb is "1% is easy, 0.1% is hard, 0.01% is very hard".
Thankfully, there are endless circuit designs which don't need perfectly matched devices because you linearise them with active loads, cascodes, parallel devices, degeneration and so forth. Work smarter not harder!
Thankfully, there are endless circuit designs which don't need perfectly matched devices because you linearise them with active loads, cascodes, parallel devices, degeneration and so forth. Work smarter not harder!
Last edited:
How about 250 of the exact same type of valve by the same manufacturer, same year, same batch codes?
Really, I spend a lot of time matching pairs and quads of valves and it works.
In fact you would be suprised just how MUCH spread there is in production.
I get a range of values from weakest to strongest on this example of 11.4m/a to 20.85m/a, so up to 80% difference.
This is normal.
I had the same story for many mass produced valves.
I have 8 boxes in front of me now from my marathon matching sessions.
This is a double matched quad to fit in an amp
date stamped all of them 26th June 1973 except one April 10 1971
Quiescent are the following values at 440V.
15.6m/a, 15.4m/a, 15.9 m/a, 15.6m/a, 15.85m/a, 15.5m/a, 15.5m/a, 15.95m/a
Ie. max spread 3.6%.
When averaged out between strongest & weakest it drops to half that so within 1.8%.
It's hard to get resistors and capacitors that close, it costs money. 🙁
I am well aware, this kind of effort is worthwhile, which is why I also match double triode pairs in the same way.
Watching them age, so long as no massive overloads take place, I haven't noticed them drift very much, and they invariably give a linearity/performance which mass produced amps can't get even close to.
I guess that was the signature of Swiss companies like Studer?
Have you ever seen inside a STUDER valve amplifier? It's a thing of beauty.
Their analogue mixing consoles were the same, like their famous Revox tape decks.
ie.-
When you watch scope traces carefully,- (I also test stuff at full power), you can spot a weak valve very easily by an assymetric trace visibly on one output phase.
You may also be able to trace assymetry in a NFB loop the same way.
On one amp I used 2 NFB loops - from the 140V CT winding to each of the phases of the phase invertor precisely for that reason.
It made a heck of a difference.
I could say, some of the last TV valves had matching sections many times better than all those overpriced wonders you see being flogged in many cases for 100s of USD.
Btw:-
I worked a long time with racing engines.
You would be suprised just how similar the procedures are.
When you start to demonstrate to people just how bad production stuff is, they often they go into denial, and that's when you're working with OEM.
It's another whole story when you start working with repro crap!
If/When you do a blueprinted modified head/bottom end for a race engine, if it has flow tolerance within 2% and compression ratios within the same band, you are not talking about the same engine, and people will swear blind it's non standard!
Last edited:
I've never matched toobz. Found the odd one defective, set it aside. And for the most part got the results I wuz looking for.🙂
If we throw enough money at a project most anything is possible. I've made a point to keep a tight lid on cost. Mostly since I didn't have the Gelt to go further.
If we throw enough money at a project most anything is possible. I've made a point to keep a tight lid on cost. Mostly since I didn't have the Gelt to go further.