John Curl's Blowtorch preamplifier

[Variac]

I would regret to have been recycled without the opportunity to share more

I think I'll use that as my epitaph

 

[jacco vermeulen]

Mark,

you're welcome to donate your remainders to the Donner culinary society.
recipes

 

[Nelson Pass]

Grey, let us separate local from global loop feedback. Black did NOT invent local feedback, they MUST have known about it for decades before he wrote his paper. When MOST engineers say negative feedback, they mean feedback over several stages. Technically, this is incomplete, but so what? It is shorthand to describe a technique. Almost everyone uses SOME local feedback, AND it has real potential problems also, BUT not as many as does GLOBAL negative feedback. That is what we have found.

Exactly right, John.

Taking this a step further, Bob did some simulations a while
back which implied that degeneration creates the same changes
in the spectral distribution of distortion as loop feedback.

I thought to myself, this cannot make sense. As I am fond of
repeating, the transistor does not know how it is used, and if
loop feedback is not being used to alter the input signal, how can
the transistor respond with a different set of harmonics? Of course
it does respond to the changes in Vds as well as Vgs.

So I build a test setup with a JFET cascoded so as to minimize the
effects of fluctuating Vds, and measured the spectral distribution
of the distortion vs degeneration. I did not find the increase in
3rd and higher harmonics relative to 2nd.

So I tentatively conclude that Bob's simulation was flawed by
not fixing the Vds value, and that loop feedback is not the same
as degeneration.

Perhaps someone will care to simulate this and see if it agrees
with what I saw.

 

[john curl]

Nelson, local feedback WILL create higher order harmonics, BUT they are in exactly the same phase/time relationship as the original distortion. This added distortion can still be important in RF input stages, but it seems that the 90 degree rolloff within the global feedback loop is the real problem, rather than a slight increase in higher order components.
Just look at a fet differential pair, all by itself. The individual fets may have mostly 2'nd harmonic when used alone, BUT when they are used in a differential pair, you get more 3'rd harmonic than either device should have, if used alone.

 

[PMA]

There is too much voodoo in this thread, most of the reasoning. It surprises me a bit.

 

[GRollins]

And if you cascode a differential...then what?

Grey

 

[john curl]

It doesn't fix the problem completely.
Think about a single mos or jfet. Bias it class A. Measure its distortion, note order of harmonics and amount.
Add a reasonable size source resistor. Bias the fet again to exactly the same current. Measure its distortion and harmonics in this case as well. What is the difference?

A diff pair has already made 3rd harmonic from the second, so it is less sensitive to this test.
However, RF amp designers take this VERY seriously. WHY?

 

[anatech]

Hi Nelson,

I would regret to have been recycled without the opportunity

Thank you for that. 😀

So I tentatively conclude that Bob's simulation was flawed by
not fixing the Vds value, and that loop feedback is not the same
as degeneration.

I don't know what happened with his experiment, but I have to agree with you. Degeneration will always reduce the amount of distortion an active device has (thinking emitter / source resistor). The higher impedance of the collector / drain may have allowed another factor to come into play.

Hi Grey,

I'm not aware of any reason why you would be bothered by the circuit.

Well, you have me dead to rights there! I'm not in the least, but it's agreed that there are those who would just love to dive into that one!

And if you cascode a differential...then what?

I imagine you fix the drain voltage and greatly reduce the distortion caused by a varying drain voltage. Your supply noise rejection also goes way up. Therefore, a good thing? This works wonders in a power supply regulator. A massive improvement in performance when used to feed your reference and output circuit assy.

Hi John,

Just look at a fet differential pair, all by itself. The individual fets may have mostly 2'nd harmonic when used alone, BUT when they are used in a differential pair, you get more 3'rd harmonic than either device should have, if used alone.

Isn't that due to the cancellation of the second harmonic as a characteristic of a differential pair? The same holds true for BJTs.

-Chris

 

[MikeBettinger]

And if you cascode a differential...then what?
Grey

You end up with yet another set of numbers that have little relationship to the actual performance of a circuit handling a signal that is simultaneously comprised of the audible bandwidth (if properly implemented) and can span many decades of the DB scale. Did I mention all at the same time? Yes I think I did.

Listening to music ROCKS!

Sorry, the voices in my head couldn't stop me this time...

 

[john curl]

No Chris, it is MORE than that.

 

[anatech]

Hi John,
Where should I look to learn more on this? I'll believe you on that, so I'm willing to study.

-Chris

 

[john curl]

This is a tricky subject, which is grad school in complexity.
A first reference is : 'Analog Integrated Circuits for Communication' Don Pederson, Kartikeya Mayaram, Kluwer Academic Publishers, 1991 pp. 106-114, for a start. Then you have to develop it for differential bipolar and finally fet diff pair.

 

[Nelson Pass]

Nelson, local feedback WILL create higher order harmonics, BUT they are in exactly the same phase/time relationship as the original distortion.

Hmmm... I was referring only to resistive degeneration with a
constant Vds. Under those conditions I am not seeing that.

 

[john curl]

Where does the distortion come from in the first place?

 

[anatech]

Hi John,

Where does the distortion come from in the first place?

I'd guess from the basic non-linearity of the active device in question. Now the trick is to pick which mechanism is / are the dominant one(s). I'm assuming that the resistor is linear enough so that it's contribution is negligible and that we are not including any problems associated with coupling capacitors.

So, are you trying to say that the differential connection is problematic? Would this not depend on the method used to generate the tail current?

This is a tricky subject, which is grad school in complexity.

Well, I did attend Rye-high (Ryerson in Toronto).

A first reference is : 'Analog Integrated Circuits for Communication' Don Pederson, Kartikeya Mayaram, Kluwer Academic Publishers, 1991 pp. 106-114, for a start. Then you have to develop it for differential bipolar and finally fet diff pair.

This would seem to be a waste of time and may possibly generate wrong conclusions on my part. Working from first principles is something that you do when blazing a path or studying math (I'm terrible with math). There must be a more sane way to study this topic. I still have some of my text books on semiconductor theory, but I doubt they would go into what looks like a third or fourth order semiconductor model. Maybe they will.

Can you give me the name of the effect or process I should be looking up? No, I don't know what you are talking about. Anyone else who does is welcome to show me the light.

-Chris

Edit: John, I see that you have pointed me to a chapter entitled "Distortion in Feedback Amplifiers". This implies that you are now looking at a loop feedback issue, rather than a local feedback phenomenon that I thought Nelson was talking about.

 

[anatech]

Hi John,
Thank you for the referral. I just ordered that book and should have it in a month or two (depending on USPS). Darn thing is expensive, but then they all are these days. It looks like an interesting book.

I was able to find a used copy for a little over 1/5 the new price. Thank goodness!

-Chris

 

[john curl]

The REAL subheading is: The C-E State with Emitter Feedback. This is ONLY the bipolar example, but it should work for fets also.
I can't do the whole thing, but it starts: "...a common emitter stage is shown including an external resistor Re in series with the emittter. Negative feedback is produced by the series {feedback) resistor. "It does a lot of heavy modeling and math and shows the cancellation of the intrinsic 3rd harmonic generated by the Gm change with current with the 3rd harmonic generated by the resistor interacting with the signal. It is presumed to be a perfect resistor, of course. Why does this happen. Why can't we cancel all 3rd harmonic as easily as this, and is it practical to do so? Hint: Think 180 degrees phase shift of the same harmonic number.

 

[john curl]

Before some of you might get even more puzzled about what I am implying, think about distortion products and that they actually have a phase shift relative to each other and to the fundamental signal. We usually ignore it, BUT it can be interesting when trying to understand local feedback.
Most 'linear' systems run out of room at the maximum of their signal swing. IF this is a really soft overload,and is symmetrical, it is primarily COMPRESSIVE 3'rd, although there might be a few more odd harmonics, if you keep pushing harder. However, there is also EXPANSIVE 3'rd harmonic which is 180 degrees out relative to compressive 3'rd. With large values, expansive 3'rd will look like a TIP at the top of a sine wave fundamental.
It just so happens that bipolar transistors have expansive 3'rd. It is in their nature. However, local feedback (or any negative feedback for that matter) tends toward generating extra compressive 3'rd, and if you are careful, you can cancel the two distortion products.
I don't think this can be done, however, with a diff pair, and it is a fundamenal weakness of this topology. The math would be formidable, but note that a bipolar diff pair always has COMPRESSIVE 3'rd. I suspect that a fet pair does as well. Where did this distortion come from?

 

[john curl]

Chris, and others concerned here, I was also surprised by the complexity of interactions with both local and loop feedback with even simple active devices. It IS math intensive,(to really understand it), but many here can probably simulate something with SPICE and see the same thing with the computer. Measurement has other difficulties, and the diff pair is even more difficult, because the 3'rd harmonic present initially, even without resistive degeneration is mostly from the interaction of the two devices, and not from the individual devices.
There was a very heavy math analysis of input stages in the IEEE Transactions a few years ago. It 'proved' that 'fet' diff pairs had more distortion than bipolar diff pairs in some realistic situations. I can't find the article or its title at the moment.

 

[anatech]

Hi John,

The REAL subheading is: The C-E State with Emitter Feedback.

Thanks. I was reading an excerpt from the book, so the information was incomplete to put it mildly. This was enough to convince me that I might get something from reading this book. The primary reason I bought it was that you had recommended it past the excerpt.

Math. Math is a short hand way of explaining something. The terminology is the hardest part of using math. I do "get" concepts and understand them well, without the math. When an author dives right into math, I'm lost. Stanley Lipshitz generally lives on the math plan and loses most other audio people. I haven't been to an AES meeting now for well over 10 years as a result of this.

My thoughts on differential pairs lead me away from using BJTs. They do not handle line level signals very well unless they are heavily degenerated. For some reason, I find that JFETs and Valves sound much cleaner. My own belief is that devices with lower slope transfer curves are better in this application. BJTs are about the highest, being exponential.

-Chris