harmonic cancellation --
Its another way than thru linearization only -- harmonic cancellation. BTW -- its mostly 2H it is canceling.
My question for all is mostly about what is actually going on that allows for cancellation this way. It is fundamental to getting best performance with minimum parts... the most elegant design approach IMO. I am sure by cascoding the jFET etc it can be reduced further, but that isnt the point of the illustration.
Thx-RNMarsh
The idea to try this came from making oscillators have low distortion.
Its another way than thru linearization only -- harmonic cancellation. BTW -- its mostly 2H it is canceling.
My question for all is mostly about what is actually going on that allows for cancellation this way. It is fundamental to getting best performance with minimum parts... the most elegant design approach IMO. I am sure by cascoding the jFET etc it can be reduced further, but that isnt the point of the illustration.
Thx-RNMarsh
The idea to try this came from making oscillators have low distortion.
Last edited:
Harmonic cancellation isn't necessarily an accurate term. My buffer cancels all harmonics because it applies a reverse matched transfer curve to the output transistor. This is different than just canceling H2. A good metric for the quality of a null is how large H3 is once H2 has been trimmed out. That's usually not much different than specifying the minimum THD figure achievable with trimming.
Isn't harmonic cancellation a method of linearization? It is in the opposite direction from feedback, but still a method of linearization.
Isn't harmonic cancellation a method of linearization? It is in the opposite direction from feedback, but still a method of linearization.
You claim your Hfe argument means my evil Ccb argument is not valid. I think it is your turn to do some sims taking at least as much care as I have to remove other effects to 'prove' this
I suspect what you will find is that .. (duu.uuh! Better wait and not anticipate what he finds )I brought up GG Baxandall's way of looking at impedances within a feedback loop really to point out that nothing is ever pure current or voltage driven. Quite often, it is very much NEITHER.
The enhancer doesn't 'voltage' drive the main VAS device. It just drives it from a stiffer source than the IPS/CM.
The IPS/CM doesn't 'current' drive the enhancer .. unless the enhancer has such poor Hfe that its input Z is much less than the IPS/CM output Z. In real life, the 2 are in the same ballpark.
Of course if there is some important advantage in some situation, you strive to make that point more current/voltage/bla bla driven etc which is the basis of my 'pure Cherry' obsession
What I was demonstrating is how the enhancer helps the amp 'ignore' modulated Ccb of the main VAS device. I expected the improvement will be greater at HF and that it would affect mainly the 2nd.
I'm slightly surprised the improvement goes down as far down as it does.
The corner for C7 is set just above 200kHz as I said in the original post. The THD analyzer does 10 harmonics so this is the limit for 20kHz THD.
Its the bare minimum to make the amp stable and show no peaking below 200kHz. Remember both amps are operating at nearly Open Loop. The idea is to have the same Loop Gain for both examples but this is impossible for flat response.
Instead we note the differences in Loop Gain, 10dB @ 40kHz dropping to none at 200kHz and see if the difference in THD is less than this. If the difference was due solely to Loop Gain above 20kHz, we would expect to see less than 10dB difference in 20kHz THD. Instead we see about 15dB improvement @ 20kHz.
ie Distortion is reduced by quite a lot MORE than the difference in Loop Gain.
So I hope you accept the converse corollary too
As I said, the practical point is that the VAS enhancer, provides MORE THD reduction than you'd expect from the increase in Loop Gain. I've proposed my main reason for this. Your turn to propose (and show evidence of) a better one if you don't like or believe mine.
_________________
In fact it INCREASES spikiness
Last edited:
Well it's clear to me that the resistor is changing the transfer function of the upper Jfet to match it better to the lower, which results in a 2nd harmonic null. The 3rd harmonic will only cancel to the degree that the transconductance curves have a mathematical aptitude, which may be more true for Jfets than BJTs.
H3 could potentially be trimmed in that circuit by putting a trimmer between the Jfet sources. After trimming H2, adjust the trimmer a bit and see which direction brings H3 down. You'll have to re-trim R4 while you do this, but you may find a setting where both H3 and H2 are cancelled, resulting in even lower THD.
But that's only if H4 or H6 doesn't end up becoming the dominant distortion, which will happen if the transconductance curves don't have that inherent aptitude for a deep null.
H3 could potentially be trimmed in that circuit by putting a trimmer between the Jfet sources. After trimming H2, adjust the trimmer a bit and see which direction brings H3 down. You'll have to re-trim R4 while you do this, but you may find a setting where both H3 and H2 are cancelled, resulting in even lower THD.
But that's only if H4 or H6 doesn't end up becoming the dominant distortion, which will happen if the transconductance curves don't have that inherent aptitude for a deep null.
Last edited:
from #4016
Schoeps use it to good effect in their ubiquitous FET phase splitter. A true Schoeps mike is set up with a THD meter. The copies don't bother. There was a discussion of this recently in MicBuilders.
Scott Wurcer deals with this in his 2 Linear Audio articles on capacitor mike head amp design.
I'm sure I saw an EE Design (??) article on this too.
This is a known phenomena.
Schoeps use it to good effect in their ubiquitous FET phase splitter. A true Schoeps mike is set up with a THD meter. The copies don't bother. There was a discussion of this recently in MicBuilders.
Scott Wurcer deals with this in his 2 Linear Audio articles on capacitor mike head amp design.
I'm sure I saw an EE Design (??) article on this too.
No I didn't. I only claimed that an enhancer itself does not increase open-loop linearity.
I brought up GG Baxandall's way of looking at impedances within a feedback loop really to point out that nothing is ever pure current or voltage driven. Quite often, it is very much NEITHER.
The enhancer doesn't 'voltage' drive the main VAS device. It just drives it from a stiffer source than the IPS/CM.
The IPS/CM doesn't 'current' drive the enhancer .. unless the enhancer has such poor Hfe that its input Z is much less than the IPS/CM output Z. In real life, the 2 are in the same ballpark.
Of course if there is some important advantage in some situation, you strive to make that point more current/voltage/bla bla driven etc which is the basis of my 'pure Cherry' obsession
What I was demonstrating is how the enhancer helps the amp 'ignore' modulated Ccb of the main VAS device. I expected the improvement will be greater at HF and that it would affect mainly the 2nd.
None of this was ever in question.
Instead we note the differences in Loop Gain, 10dB @ 40kHz dropping to none at 200kHz and see if the difference in THD is less than this. If the difference was due solely to Loop Gain above 20kHz, we would expect to see less than 10dB difference in 20kHz THD. Instead we see about 15dB improvement @ 20kHz.
ie Distortion is reduced by quite a lot MORE than the difference in Loop Gain.
So I hope you accept the converse corollary too
As I said, the practical point is that the VAS enhancer, provides MORE THD reduction than you'd expect from the increase in Loop Gain. I've proposed my main reason for this. Your turn to propose (and show evidence of) a better one if you don't like or believe mine.
Again, this is not in question. The point I was actually making is supported by your simulations already in a detailed way. Notice that at 20KHz, H3 IS reduced by 10db. H2 is the exception. It is reduced by 2 times more than H3. If there were not some counter-H2 going on, then H2 would be reduced by only 10db as you would expect. So your enhanced TIS is ultimately less linear, but global linearity is increased simply because it cancels the H2 of a different stage.
Err..rh! H3 @ 20kHz is 60kHz and the difference in Loop Gain is LESS than 10dB. The difference in Loop Gain at 40kHz, H2 is 10dB and again the reduction is a lot more than that.
There are 2 possible explanations.
- One is that the enhancer has 'linearized' the VAS (or some other stage).
- The other is that it is doing 'distortion' cancelling' of the VAS (or some other stage).
__________________
But what is the purpose of this discussion? Are you proposing a new distortion reduction mechanism to explain my results? That would at least be useful. I would certainly be eager to see this.
In the end, the exact mechanism is only important if we have to do something different to take full advantage of this.
My main interest has always been how to get supa dupa performance from naive & mundane circuits. For me, the subtle 'advantage' of the humble VAS enhancer is exactly the type of stuff I'm always looking for.
If you think some other mechanism is responsible for the 'better than expected' results, please explain it (and show some careful evidence) so we can take better advantage of it.
Last edited:
Well, that is side-stepping the answer. I am sure this is known by someone before. Is basic fundamentals which seem to get lost now and then.
It is put up here to illustrate a pure harmonic cancellation technique that does not rely on open loop/closed loop gain et al. But if the understanding of why this happens is used along with other more commonly known methods to reduce distortion - NFB etc)... we could get some impressive numbers with much less effort.
[btw - this little buffer experiement has its H2 and H3 well below -120dB. Noise limited test.]
Thx-RNMarsh
Last edited:
In my mind there is still a distinction. So I think harmonic cancellation is not linearizing. Cascoding - now that is linearizing. But it isnt that important, IMO. Just different ways to reduce the distortion at the output.
Thx-RNMarsh
Like I said, distortion cancellation is a method of linearization. So you're not arguing with me. As I said,
That is to say, merely the function of providing current gain or changing impedance does not increase open-loop linearity (which is independent of loop gain). However the harmonics introduced by the enhancer may cancel harmonics coming from elsewhere. The point is that the mechanism causing greater open-loop linearity is harmonic cancellation, and that is the answer I gave because you asked the question.
Like I said I will not do it today because I need sleep, but unless you feel my simulations will be an improvement over yours, I suggest you compare the FFT or .four result of the common-emitter base current and the beta-enhancer base current. If I'm right, you will see that the Ib of the enhancer has a smaller H2 than the common-emitter Ib, but not for the 3rd harmonic.
The fact that the cancellation follows from LF to HF means that it is not the Ccb of the TIS that is causing the distortion, but the loading of the TIS by the Ib and Ccb of the drivers. Otherwise the H2/H3 ratio would change because the cancellation would break down as common-emitter base impedance changed with frequency. If I am correct, it should not, at least within the audio spectrum.
Last edited:
Nonlinearity just means deviation from a slope. You can have a spiky transfer curve and smooth transfer curve and they will both have the same nonlinearity, just a different distribution of harmonics. Whether you reduce the harmonics by increasing the slope (degeneration), applying feedback, or by canceling a harmonic, the effect is the same - to make the circuit more linear. The distinction is that degeneration and feedback essentially swamp the distortion to reduce it to a number of zeros, whereas cancellation does this by directly subtracting the distortion. It is a striking difference but does not make cancellation any less a method of linearizing the circuit. We just do not usually think of anything but feedback as linearizing.
You need to speak for yourself, these techniques are well documented from at least 1968 or earlier.
come on, guys. Details. give the exact reason... not a generalization on transfer curves better matching.
THx-RNMarsh
He did. The Vgs vs Vds is an empirical trancendental function, the x^^2 terms can be made to cancel by an empirical tweak. This point is quasi-stable and I doubt -120dB would hold over time and temperature, you already show it does not hold for supply variation.
Last edited:
Thats a little closer to what i was looking for. But even more to the point than Vgs/Vgd is the C of the devices changes and is shifted (by R4) to match the complimentary device. Getting the C's equal makes it work.
___________
The same null can be obtained by offsetting the + and - supplies for a THD null. It takes several volts offset with these transistors - more than a regulated equal volt supply would move. But it is shown that way the C changes enough to find a match with the compl side.
Another way to use this to usefull effect is a single jFET, common source - the C multiplied by the gain is a larger value than the input C. If you added C on the input to equal the effective C at the output, the two are 180 degrees out of phase and you get a THD null. At least to the first order and that can be a lot. [ of course a series buffer r on the gate before the added c is needed]
-THx RNMarsh
___________
The same null can be obtained by offsetting the + and - supplies for a THD null. It takes several volts offset with these transistors - more than a regulated equal volt supply would move. But it is shown that way the C changes enough to find a match with the compl side.
Another way to use this to usefull effect is a single jFET, common source - the C multiplied by the gain is a larger value than the input C. If you added C on the input to equal the effective C at the output, the two are 180 degrees out of phase and you get a THD null. At least to the first order and that can be a lot. [ of course a series buffer r on the gate before the added c is needed]
-THx RNMarsh
Last edited:
DC or AC? The DC null is all I was talking about. The zero TC point has nothing to do with the capacitances. The Schoeps circuit mentioned was one FET in any case where it was DC loadline manipulation, the capacitive portion was small.
I checked your post again at 1kHz you show the difference as almost three orders of magnitude from 12V to 22V supplies. Or am I reading the wrong line?
Last edited:
I don't follow your added text. The noise of any gate input r is unacceptible. The circuit mentioned was tuned by walking the load line until the output distortion cancelled the inherent squared term in the Vgs (forward gain) of the amplifier. It had nothing to do with capacitances, so we are talking about two different things. Nelson Pass has also used this.
The Scheops circuit is a phase splitter with equal 2.2K source and drain resistors, there is no gain in this circuit (except the +2 for being differential). In any case here the capacitive non-linearity (are you even talking about the voltage coefficient of the junction caps?) is in quadrature with the forward gain of the amplifier and can not null. The first time I discussed this somene thought balancing the gain of the phase splitter cancelled seconds, it does not.
Last edited:
Focused on the C part -
Good info. But you missed the point. I'm not inventing anything.... though i told JC this many decades ago and it appeared new at the time. At least to him. Never-the -less, I disagree -- The device C is important and part of the overall scheme of things... esp distortion caused by the C part. You can use it to cancel distortion in certain applications.
Another approach used in differential stages is to cross-couple C to cancel [or at least reduce] the C's and increase the stage BW.
Wouldnt want some to forget or overlook opportunities with the C for ever finer tuning.
Thx-RNMarsh
Good info. But you missed the point. I'm not inventing anything.... though i told JC this many decades ago and it appeared new at the time. At least to him. Never-the -less, I disagree -- The device C is important and part of the overall scheme of things... esp distortion caused by the C part. You can use it to cancel distortion in certain applications.
Another approach used in differential stages is to cross-couple C to cancel [or at least reduce] the C's and increase the stage BW.
Wouldnt want some to forget or overlook opportunities with the C for ever finer tuning.
Thx-RNMarsh
Last edited: