The Sound of Science

[jan.didden]

So, an inverting topology like folded cascode should help with this issue ?


Mike

What would help is a more linear forward gain path. Does FC help here? I think it does.

Jan Didden

 

[MikeB]

Jan, of course FC helps with linearity, but my idea was that the
error introduced in the "first" loop, is negated in the 2nd loop
of the fb, so cancelling errors out instead of adding them...
(I know that there is no such thing like first loop in fb, was just easier to explain this way.)

Mike

 

[jan.didden]

Jan, of course FC helps with linearity, but my idea was that the
error introduced in the "first" loop, is negated in the 2nd loop
of the fb, so cancelling errors out instead of adding them...
(I know that there is no such thing like first loop in fb, was just easier to explain this way.)

Mike

Yes you are right, and in the end it gives you a more linear "black box" amplifier. I'm sure David will like that analogy

Jan Didden

 

[lumanauw]

are you suggesting that there exists a (finite) threshold speed above which transistors can eliminate (as in complete, 100% reduction of) higher order harmonics?

Eliminate ? (I'm sounding like MikeB )

As long as you don't have "linear magnification" transistor, you will never ever get rid of higher order harmonics.

This quote from "Mr Black Box" himself is very good :

The transistor has a non-linear transfer characteristic. Initially, amplifying a signal, this will cause harmonics and intermodulation (depending on the input signal). Because the feedback returns those harmonics to the input, they are again processed by the same non-linear characteristic that caused them in the first place. So you get harmonics-of-harmonics and intermodulation products of intermodulation products. That causes the increase in higher harmonic (and dis-harmonic) products. The transistor remains the same, it is still the same process that does it. There is no trading and nothing magic because of feedback happens to the transistor.

Don't forget that whatever you do to the loop, the amp itself will ALWAYS work open loop. If you have an amp with 1000 gain OL, and you apply feedback to get the gain down to 10, the amp still needs the input signal for its 1000 x gain. Assume you have 10V Vout, the effective amp input voltage is 10mV. This does not change whether you use fb or not. With fb and CL gain of 10 x, the effective Vin is comprised of 1 V input signal (to get 10 V out with 10 x CL gain), and 0.99 V fb signal. The effective 10mV input signal will look very distorted (because it is Vout with Vi subtracted). That is kind of "pre-distorted" so the Vout will look almost undistorted.

If you want to improve it, use more linear transistors (or topologies) and/or use faster transistors so the phase shift is less so the error cancellation by the fb is more accurate (see my earlier post).

Here's a clue. Why if we use FET as differential, it sounds better? Is it because fet can handle more input voltage (Bipolars W/O degeneration only needs few mV to saturate)? No, I don't think so. If the criterion is big input voltage that can be handled, then bipolar differential built with quite some RE will be the same.

The key is not how big input that can be handled. The key is how near the magnification curve to the ideal "linear from 0" curve.
Bipolars have exponential curve. FET has square law curve. Square law curve will always "near-er" to linear curve.

How to make (if you want to) 2nd harmonic? Easy, just "unbalance" the current sharing between left and right transistor of the differential pair. If the ccs is 2ma, the perfect 1ma-1ma sharing will be the most clean one. If you set the collector resistor so the left is 0.2mA and the right collector is taking 1.8mA, you will be able to hear what the 2nd harmonic sounds like. (this is what I see in FM Accoustic)

Bipolars will exhibit "easier" to get more 2nd harmonic by that way, because its magnification curve is exponential (compared square law of FET). Don't forget if it is easier for 2nd harmonic, it will also be easier for creating 3rd-4th-5th.....etc.

So, bipolars will be more prone to producing higher order harmonics than FET, for the same cause. It is harder to make higher order harmonics from square law magnification curve, compared to exponential magnification curve.

With FET, usually we get more "pleasant" sound than if we use bipolars for the same topology, because there will be less higher order harmonics from the square law transfer curve (compared to exponential curve).

 

[MikeB]

lumanauw,
"square"-transfer curve (parabel,x^2) only creates 2nd harmonics,
in combination with a diffamp, 3rd harmonics only. (2nd canceled out when balanced)
Sadly, jfets have more a x^3 function, but still far superior to bjts.
The only devices i know of giving a nearly pure x+x^2 curve are tubes...

But, it seems, the worst harmonics are created in the outputstage...

Mike

 

[darkfenriz]

With JFET, usually we get more "pleasant" sound than if we use bipolars for the same topology, because there will be less higher order harmonics from the square law transfer curve (compared to exponential curve).

I have my own (???) reasoning why JFETs often sound better in input LTP. It is a matter of transconductance.
Assume you have 'normal' hi-gain class AB amp with bandwidth 50-150kHz and open loop bandwidth 5-10kHz. Its open loop sensitivity is less than 1mA. Now, output creates wide band Xover distortion with 300kHz components of , say, 1mV. This 1mV of 300kHz is fed back to input LTP, with bipolars 1mV of Vb-e difference may cause non-linear behaviour (nearly swith-off or too high current), because of high transconductance, with Jfets rather not. Worst things happen if LTP bipolars have no emitter reistors and are current mirror loaded. Interesting, that nothing can filter out such HF components before VAS dominat pole (often Miller cap).
regards

 

[MikeB]

I do not agree with you about the transconductance...
If it would be like that, a simple RE in ltp would make sound bjts
like jfets... And, the sk170's transconductance reaches that of bjts.

But, i agree with the saturation caused by hf-content from xover-distortion,
if it does saturate the inputstage, you would get plenty of TIM.

Mike

 

[lumanauw]

Hi, MikeB,

So it is reasonable that tubes sound nice? It's hard to make high order harmonics with tubes (because of the nature of its magnification curve), right?

Hi, Darkfenriz,

I always curious about how the transfer curve from a CFP bipolar transistor. Is it linear from 0 or not linear at all?

What do you think of "CFP everywhere" audio power amp (I think MikeB has tried this?). The differential is CFP bipolar, the VAS is also CFP bipolar, the final stage is also CFP bipolar, what it will sounds like?

 

[jcx]

Unfortunantly degenerating threads don’t improve content, or even keep up with what has been discussed elsewhere

But degenerating bjt’s gives better than fet results as I have shown:

http://www.diyaudio.com/forums/showthread.php?postid=501789#post501789

further the principle of higher harmonic multiplication by feedback can be misleading, the practice shows that enough loop gain and high feedback (+100dB loop gain over the audio range isn’t that challenging) can drive even the “multiplied harmonics” down below the noise floor:

http://www.diyaudio.com/forums/showthread.php?s=&postid=697517&highlight=#post697517

a further diversion you seem to be having trouble with is the performance goals of audio amplifiers – the dominant “deal” is for audio amp designers to deliver low noise, flat frequency response and small output impedance (high feedback makes it easy to deliver on all of these over the audio frequency range – output Z at the amp terminals can be orders of magnitude lower than any reasonable speaker cable)

you can prefer other responses from your amplifier but I contend that frequency response eq of your source and tailored, frequency dependant damping for this or that speaker is not the goal of a “perfect” audio amplifier – you should apply eq up front and the speaker/crossover designer should have dealt with electro-acoustic interactions assuming a near zero amp output impedance – starting from zero impedance from the amplifier he can add whatever passive network required as part of the crossover design

in diy you can of course tailor your system by deviating from the goal of a “neutral” amp but it would be far easier and more transferable to others if you designed a neutral amp and trimmed your response elsewhere

 

[jan.didden]

jcx,

I agree that you can drive the higher harmonics down to insignificant levels if you make the fb high enough (high enough loop gain). I merely wanted to explain where the "new" higher harmonics come from.
Also agree with the rest of your post.

Jan Didden

 

[MikeB]

Hi jcx, i definitely agree with you that properly used feedback reduces
highorder harmonics below audibility, and yes, degenerated bjts
show better performance than jfets in sims, but in realworld the
jfets just slightly sound better... This is not about frequencyresponse or
coloring or frequencydependent damping.
Obviously there are more parameters than simple linearity and
flat frequency response defining the sound of an amp ?
Of course an audioamp should simply reproduce as exact as possible.

And about the "vasload bs", sorry, i showed you that rloading can
substitute or even outperform re's...

Mike

 

[Nelson Pass]

"square"-transfer curve (parabel,x^2) only creates 2nd harmonics,
in combination with a diffamp, 3rd harmonics only. (2nd canceled out when balanced)
Sadly, jfets have more a x^3 function, but still far superior to bjts.
The only devices i know of giving a nearly pure x+x^2 curve are tubes...

And yet I routinely see a nearly pure 2nd harmonic with JFETs
and MOSFETs in single ended Class A circuits.

 

[tlf9999]

And yet I routinely see a nearly pure 2nd harmonic with JFETs
and MOSFETs in single ended Class A circuits.

When reality doesn't match with our imagination, we will have to abandon reality.

 

[PMA]

And yet I routinely see a nearly pure 2nd harmonic with JFETs
and MOSFETs in single ended Class A circuits.

Yes. The 2nd is usually that high in this case, that one tends to overlook the higher order harmonics, that are almost always higher in magnitude compared to serious designs. It is interesting, that -120dB of say 7th harmonics of a feedback design is wrong, contrary to usual -90dB 7th harmonics of SE, which is perfect ;-). I hope you don't believe it in fact.

 

[PMA]

Here is one of those "purely 2nd harmonics":

 

[darkfenriz]

Yup, that's the point.
But concerning to whom you answer, it still looks like 'teaching your father how to make children'. By the way I suspect Mr. Pass does not share his whole reasoning about what and why he does something exactly THAT way and not any other.

 

[PMA]

Yup, that's the point.
But concerning to whom you answer, it still looks like 'teaching your father how to make children'.

I am not that young to be his son and probably old enough to be a father of most writers here, especially university students . My advantage might be that I do not make a living by audio designs .

 

[darkfenriz]

may I call you "uncle" ?

 

[anli]

OT

To PMA
Pavel, is your email (published at your site) live?

P.S. Sorry for OT

 

[ilimzn]

The transistor has a non-linear transfer characteristic... this will cause harmonics and intermodulation... Because the feedback returns those harmonics to the input, they are again processed by the same non-linear characteristic... you get harmonics-of-harmonics and intermodulation products of intermodulation products.

That is exactly the mechanism, in a nutshell.
However, I think it needs some clarification:

1) It is the non-linearity of the amplifying elemet that causes higher harmonics. Because the gain of the amplifying element is not constant with frequency, the effective feedback is not the same at all frequencies, so the harmonic spread of the distortion spectrum is not the same as would be calculated from the raw transfer characteristic (BJT, JFET, MOSFET, SIT, Tube...). In other words, because the speed of the amplifier is limited, feedback by necessity attenuates high order distortion products less than low order.
This is very important regarding the way the human ear functions - a built-in mechanism is capable of masking high order harmonics with low order, IF they are present in the right mix (research is not exact about this, but a falling series with 10-20dB less on each subsequent harmonic seems the general ballpark figure). This is where the problem lies - the harmonic series looks quite different in general FB amps. The point is, that what is ABOVE the noise floor needs to, as much as possible, follow a falling series, within the audio band (and preferably a ot wider). IMHO this is where things like SE Triode amps get their reputation, but it is also for me, a sort of trivial, or if you like, somewhat cowardly approach to the problem. These amps may well generate the proper series of harmonics and sound nice, but the point is the harmonics were not supposed to be there in the first place, so they do not sound accurate. Feedback can indeed be used to create an amp where for all intents and purposes, what is visible (if anything) over the noise floor, has the right series decay, and it invariably starts with designing a linear, very wide band amp BEFORE feedback is applied, and THAT is no small feat. It is far more an art than stringing an exotic triode to even more exotic passive parts - and I do think I am entitled to say this because I have done (and still do) both solid and vacuum - which IMHO is the reason why really only a few out of a mirriad amps sound BOTH nice and accurate.

2) The fact that the feedback signal is again processed by the same non-linear element again results in a modified harmonic spread. However, this mechanism is intrinsic to feedback on any level, and normally contributes to the harmonic spread modification far less than the gain vs frequency characteristic. This is because essentially you superimpose a symetric version of the original harmonic spread onto the original harmonic spread, by intermodulation. It is important to say that under stationary sine stimulus, this mechanism CANNOT generate non-harmonic distortion, because sums and differences of harmonically related sines are still harmonically related sines. In cases where a complex signal is fed into the amp, the recurring intermodulation looks more like a rised noise floor. Concievably, this would be seen as intermodulation components in bad cases, and slightly shaped noise in better cases, only if the amp CL gain falls abruptly AND at the same time close to the audio range of frequencies. I believe this is where the idea of TIM originates, so I'm not going to flog a dead horse here.

3) It is possible to produce disharmonic distortion in FB amps, even under perfect sine stationary condition input, but this is VERY unlikely to be seen at any level in the audio band. Consider this: feeding a perfect sinewave into a zero delay amplifier will produce a spread of harmonics of the fundamental frequency, given 1) and 2) above. However, if the amplifier does not have zerop DELAY (not to be confused with phase shift!) it's speed of transit essentially represents a time constant the inverse of which can also be expressed as an intermodulating frequency. Janneman has posted in a previous post that it is on the order of ps, but for discrete and even most IC amps I would tend to disagree, it would be more on the order of ns, sometimes tens of ns. Even so this translates to a frequency in the 100MHz range, at which also the OL gain of the amplifier is at best near 1. In other words, what you get from this is a mirror image of all the intermodulation and distortion components aound 100MHz, superimposed to the original set, BUT attenuated by the feedback loop attenuation. Lets try a bit of math: feed the amp 100kHz (well outside the audio band!) and it's 999th harmonic will be subtractively intermodulated around the 100MHz delay induced 'carrier' to 100kHz, reduced by the feedback factor. How large is the 999th harmonic? VERY small, way below noise. And attenuate this by the feedback's 20-100dB further... I think the point is made. Still, immagine you have a 110kHz signal. It's intermodulated 999th harmonic would fall at 109.79kHz, so not harmonically related - but so far below the noise level it is truly negligible. It should be noted that because this is a true delay and not derived from phase shift, it does not necesairly have anything to do with the amplifiers frequency response. In fact, one could cascade many very high bandwidth amps and transmission lines, resulting in a large delay, which reciprocal would be well below the bandwidth figure. Only in such an amp would nonharmonic distortion generation by feedback (assuming it does not become an oscillator!) become a serious problem, and I dare say no audio amp is made like that.

Don't forget that whatever you do to the loop, the amp itself will ALWAYS work open loop... This does not change whether you use fb or not. With fb and CL gain of 10 x, the effective Vin is comprised of 1 V input signal (to get 10 V out with 10 x CL gain), and 0.99 V fb signal. The effective 10mV input signal will look very distorted (because it is Vout with Vi subtracted). That is kind of "pre-distorted" so the Vout will look almost undistorted.
If you want to improve it, use more linear transistors

This is another brilliant point, and also the reason why for instance degenerated BJTs do not produce the same distortion as JFETs, even when the gains ae well matched. Simply put, the distortion generating nonlinearities remain the respective same for BJT and JFET, only reduced by the FB applied to the BJT. The BJT still remains essentially exponential, the JFET essentially 3rd order.

and/or use faster transistors so the phase shift is less so the error cancellation by the fb is more accurate (see my earlier post).
Jan Didden

Though, I would say faster topologies address 1) and 2) above, while faster parts also address 3) while naturally fundamentaly related to reducing 1) and 2).