Lumba Ogir said:
Can you show me please your more balanced picture so we can see fundamental differences?
distortion shaping
A long shot:
If you look at the conceptual circuitry for digital converter noise shaping (trust me, it is on topic...) you essentially see an error correction circuit: you take the difference of the input and output of the quantizer (which is the noise) and feed it back to the input through some transfer function (the shaper).
What that does is to decrease the in-band noise, and moves that noise to out-of-band, north of the audio.
Isn't feedback a kind of 'distortion shaping'? You decrease the in-band harmonics, but they pop up again as higher harmonics, higher up in the spectrum.
For higher audio frequencies, say above 8kHz or so, you don't need 'shaping' as the harmonics are above the audible bandwidth anyway. But for low-mid frequencies, you would like the first couple of harmonics to decrease a lot, even if they pop up as much higher harmonics above 20kHz. The amount of 'shaping' apparently depends on the OL harmonic contents, the OLG and the amount of feedback/ULG, but maybe there is something to be gained by carefully shaping the OL gain, distortion and feedback to get ALL harmonics, independently of frequency, above 20kHz.
As I said, a long shot...
Jan Didden
A long shot:
If you look at the conceptual circuitry for digital converter noise shaping (trust me, it is on topic...) you essentially see an error correction circuit: you take the difference of the input and output of the quantizer (which is the noise) and feed it back to the input through some transfer function (the shaper).
What that does is to decrease the in-band noise, and moves that noise to out-of-band, north of the audio.
Isn't feedback a kind of 'distortion shaping'? You decrease the in-band harmonics, but they pop up again as higher harmonics, higher up in the spectrum.
For higher audio frequencies, say above 8kHz or so, you don't need 'shaping' as the harmonics are above the audible bandwidth anyway. But for low-mid frequencies, you would like the first couple of harmonics to decrease a lot, even if they pop up as much higher harmonics above 20kHz. The amount of 'shaping' apparently depends on the OL harmonic contents, the OLG and the amount of feedback/ULG, but maybe there is something to be gained by carefully shaping the OL gain, distortion and feedback to get ALL harmonics, independently of frequency, above 20kHz.
As I said, a long shot...
Jan Didden
Jan,
It`s exactly what it is, distortion redistributing, as distortions once occurred, there`s no way to get rid of them. Preventing would be much better, for instance, by using devices in their linear region. The main problem is, and has always been, the need for power. The currently available devices maybe are able to deliver a couple of watt with fairly low distortion, but we need hundreds of watt, which is possible at the price of higher distortion. So we are forced to look at max ratings instead of linear operating range and go for efficient instead of good sounding topologies. (This is not a comprehensive explanation).
Not at all, I find it relevant.
>Isn't it not much better to cancel even harmonics using balanced
circuits, and only optimise the load line to minimise 3rd and 5th ?
Better? I don't know, although I've done both.
>How could cascoding be employed to benefit in that sense ?
If you cascode so as to freeze the value of Vds in a square law
device such as a Jfet, you tend to suppress 3rd and higher harmonics,
and of course you can then balance such circuits.
An audio haiku:
Cascode gives second,
vanished by symmetry.
Look! Here comes the third!
😎
circuits, and only optimise the load line to minimise 3rd and 5th ?
Better? I don't know, although I've done both.
>How could cascoding be employed to benefit in that sense ?
If you cascode so as to freeze the value of Vds in a square law
device such as a Jfet, you tend to suppress 3rd and higher harmonics,
and of course you can then balance such circuits.
An audio haiku:
Cascode gives second,
vanished by symmetry.
Look! Here comes the third!
😎
> If you cascode so as to freeze the value of Vds in a square law
> device such as a Jfet, you tend to suppress 3rd and higher harmonics,
> and of course you can then balance such circuits.
And degeneration would have the opposite effect ?
Patrick
> device such as a Jfet, you tend to suppress 3rd and higher harmonics,
> and of course you can then balance such circuits.
And degeneration would have the opposite effect ?
Patrick
> And degeneration would have the opposite effect ?
With degeneration you don't have a square law device to start with. With a pure square current law and perfect balancing you can get a perfectly constant transconductance in a big region (the complete class-A region), and with a little luck depending how the square current law degenerates into something more linear you can obtain quasi-constant transconductance in class-B operation, too. Search the forum for D2S or D²S, for details
- Klaus
With degeneration you don't have a square law device to start with. With a pure square current law and perfect balancing you can get a perfectly constant transconductance in a big region (the complete class-A region), and with a little luck depending how the square current law degenerates into something more linear you can obtain quasi-constant transconductance in class-B operation, too. Search the forum for D2S or D²S, for details
- Klaus
degeneration resistors
Or use a PNP/NPN pair (CFB input stage) and ensure that RE+re=0.5*VT/Ic.
RE=external degeneration resistor
re=internal emitter resistance
VT=thermal voltage (26mV at room temp.)
Or use a PNP/NPN pair (CFB input stage) and ensure that RE+re=0.5*VT/Ic.
RE=external degeneration resistor
re=internal emitter resistance
VT=thermal voltage (26mV at room temp.)
KSTR said:
My understanding is that constant transconductance in this case is
approached as the signal level goes to 0.
😎
Re: distortion shaping
Hi Jan,
The analogy to the noise-shaping converters is a very interesting one, and perhaps intuitively appealing. However, I'm not sure to what extent it applies.
Not being an expert in the delta-sigma field, I think that the noise shaping approach largely preserves noise power and shifts it out of band to higher frequencies.
On the other hand, I don't think the application of NFB preserves distortion power and just shifts it to higher order. This is the temptation that trapped people into drawing more conclusion from the Baxandall studies than was warranted.
We also need to remember that harmonic distortion is not necessarily what we hear. Just because we supposedly cannot hear the harminics of a fundamental greater than 10 kHz does not mean it is OK to have lots of 10 kHz THD. The key is IM distortion. Often the spit you hear in a cymbal is due to high-frequency IM products that land at lower frequencies.
We must always remember that harmonic distortion is a symptom, not a distortion mechanism. The same distortion mechanism that creates what we measure as THD also creates what we measure as IM. It is not completely unlike testing for a disease. The doctors are looking at antibodies most of the time - they are not the disease, but are created as an artefact of the presence of the disease.
This is not to say there is a one-size-fits-all distortion measurement. Different distortion measurement protocols excite the nonlinearities in different ways, and also have different measurement sensitivities to the results of those nonlinearities.
Cheers,
Bob
janneman said:
Hi Jan,
The analogy to the noise-shaping converters is a very interesting one, and perhaps intuitively appealing. However, I'm not sure to what extent it applies.
Not being an expert in the delta-sigma field, I think that the noise shaping approach largely preserves noise power and shifts it out of band to higher frequencies.
On the other hand, I don't think the application of NFB preserves distortion power and just shifts it to higher order. This is the temptation that trapped people into drawing more conclusion from the Baxandall studies than was warranted.
We also need to remember that harmonic distortion is not necessarily what we hear. Just because we supposedly cannot hear the harminics of a fundamental greater than 10 kHz does not mean it is OK to have lots of 10 kHz THD. The key is IM distortion. Often the spit you hear in a cymbal is due to high-frequency IM products that land at lower frequencies.
We must always remember that harmonic distortion is a symptom, not a distortion mechanism. The same distortion mechanism that creates what we measure as THD also creates what we measure as IM. It is not completely unlike testing for a disease. The doctors are looking at antibodies most of the time - they are not the disease, but are created as an artefact of the presence of the disease.
This is not to say there is a one-size-fits-all distortion measurement. Different distortion measurement protocols excite the nonlinearities in different ways, and also have different measurement sensitivities to the results of those nonlinearities.
Cheers,
Bob
Hi Nelson,
Maybe you recall Ian Hegglun's D2S article which was discussed here and which cites an article by M.Williams, "Making a Linear Difference to Square-Law FETs", EW+WW, Jan. 94, p. 84
See also attached figure from the Hegglun article. Key point is fig 1c, note the bias current is 1/4 of the total class A current as the main design criterion. When the devices go to linear gain instead of quadratic at just that point, class-B operation is possible. Of course all that is the simplified view, real FETs won't behave this ideal etc etc. IIRC, PMA wrote he tried the original Hegglun concept(s) and ran into issues.
- Klaus
Maybe you recall Ian Hegglun's D2S article which was discussed here and which cites an article by M.Williams, "Making a Linear Difference to Square-Law FETs", EW+WW, Jan. 94, p. 84
See also attached figure from the Hegglun article. Key point is fig 1c, note the bias current is 1/4 of the total class A current as the main design criterion. When the devices go to linear gain instead of quadratic at just that point, class-B operation is possible. Of course all that is the simplified view, real FETs won't behave this ideal etc etc. IIRC, PMA wrote he tried the original Hegglun concept(s) and ran into issues.
- Klaus
Attachments
There is no doubt that we can get very good linearity. If you would
care to drop the "perfect", we would be in agreement.
😎
care to drop the "perfect", we would be in agreement.
😎
Jan, “noise shaping” is a consequence of the loop gain in delta-sigma circuits, you should be able to pretty much read off the loop gain (technically the “desensitivity”) as the inverse of the noise amplitude – in fact they have some of the consistently highest feedback factors over the audio range of any audio circuitry
However the distortion reduction by feedback can be even larger – in the instance of input tanh distortion high loop gain after the diff pair reduces the signal at the diff pair giving n-1 power of the loop gain reduction of the distortion generated in a n_th order nonlinearity, literally “linearizing” the input stage
and then loop gain reduces the effect of the generated distortion by an additional factor of loop gain for a combined distortion reduction factor of the n_th power of the loop gain for underlying n_th order nonlinearity – 20 dB global loop gain increase reduces the diff pair tanh distortion contribution to the output by 60 dB (for the dominant 3rd order nonlinearity)
Cherry, “Estimates of Nonlinear Distortion in Feedback Amplifiers” JAES V48#4 2000 shows this facet of feedback distortion reduction
there is a “conservation” relation for the total amount of feedback that can be applied – look up “Bode Integral”
http://trs-new.jpl.nasa.gov/dspace/bitstream/2014/19495/1/98-0905.pdf
However the distortion reduction by feedback can be even larger – in the instance of input tanh distortion high loop gain after the diff pair reduces the signal at the diff pair giving n-1 power of the loop gain reduction of the distortion generated in a n_th order nonlinearity, literally “linearizing” the input stage
and then loop gain reduces the effect of the generated distortion by an additional factor of loop gain for a combined distortion reduction factor of the n_th power of the loop gain for underlying n_th order nonlinearity – 20 dB global loop gain increase reduces the diff pair tanh distortion contribution to the output by 60 dB (for the dominant 3rd order nonlinearity)
Cherry, “Estimates of Nonlinear Distortion in Feedback Amplifiers” JAES V48#4 2000 shows this facet of feedback distortion reduction
there is a “conservation” relation for the total amount of feedback that can be applied – look up “Bode Integral”
http://trs-new.jpl.nasa.gov/dspace/bitstream/2014/19495/1/98-0905.pdf
Re: Re: distortion shaping
Hi Bob,
Yes, that is an important distinction; noise shaping preserves the noise power, only moves it out of band. Not so with distortion reduction by fb. Thanks.
Is there any literature you (or anybody else) is aware of that discusses the tradeoffs between IM and THD? I mean, how to manipulate a design so that you can have on one side of the spectrum (no pun) lower THD and higher IM, or vice versa? I understand your note about both being a symptom for the 'disease', but I assume that depending on the desing particulars, one symptom may be more expressed than the other.
jd
Bob Cordell said:
Hi Bob,
Yes, that is an important distinction; noise shaping preserves the noise power, only moves it out of band. Not so with distortion reduction by fb. Thanks.
Is there any literature you (or anybody else) is aware of that discusses the tradeoffs between IM and THD? I mean, how to manipulate a design so that you can have on one side of the spectrum (no pun) lower THD and higher IM, or vice versa? I understand your note about both being a symptom for the 'disease', but I assume that depending on the desing particulars, one symptom may be more expressed than the other.
jd
jcx said:
jcx,
I am aware of Cherry's paper havning read it in the past, must read it again though. Memory isn't as good as it used to be 😱
Maybe I should set aside some time to look into the tradeoff between OL gain, OL linearity and amount of global feedback.
One of the positive aspects of a topology like Bob's ec amp is, for me, that the error signal handled by the global feedback loop is much reduced in amplitude, especially in the xover region, which in turn makes the input signal to the input stage much more 'linear' if you know what I mean.
You would still need the same sort of effective input amplitude because in the end it is the same OL gain that provides the output signal. So you end up with a loop being more linear and an input (effective) signal being more linear. It is possible that this reduces distortion disproportionally to what you would expect from feedback factors alone. Or maybe not. What do you think?
jd
Bob,
What happens to the distortion power according to you? It cannot just disappear, anyway.
100% agreed
Lumba Ogir said:
If you can create it, you can destroy it as well.
BTW, distortion is no such thing as mass or energy, rather information, that doesn't obey the conservation law of mass or energy.
Jan,
The industry standard 20kHz bandwidth recommendation should not be taken seriously from any point of view, it does not even cover the band of human voice, some musical instruments produce energy above 100kHz, probably much higher than that. As for the very high order harmonics, which always are followed by IM products, typically having three times higher level, no relationship with the fundamental tones, constantly changing, they do modulate the audible range (no matter if the limit of hearing is 30kHz or 10kHz). It is important to identify the enemy correctly, that is not the innocuous second and third harmonic, most often constituting a too low level for a healthy tonal balance or to be at all audible at normal listening (around 80dB) and higher SPL.
Agreed.
Lumba Ogir said:
Like material thought?
Do you remember my hybrid amp you criticized because of a global feedback around very linear tube amp and very linear class A SS output stage? What can you say now?