John Curl's Blowtorch preamplifier

[scott wurcer]

Originally posted by john curl
Scott, you are incorrect. I have seen slide rules PROVE that computer circuit analysis was incorrect. That is what gives me wisdom!

Sorry on this one you are incorrect. You're speaking out of context based probably on bad models or poorly implemented convergence criterion. I simply stated that a simple system of simultaneous equations has a unique solution in the general case. Certainly the the five or six node complementary diff-amp when stated as a matix problem with simple square-law and lambda JFETs will have the same answer. Try some time to invert a 6x6 symbolic matrix, you will see why why not just SPICE but MATLAB and MATHCAD exist.

But please elaborate, I learned to solve simple 1-2 transistor circuits with a sliderule. Solving three equations for three unknowns, for instance, pain in the butt sometimes. Typing the same matrix into MATHCAD, I usually trust the answer.

Or I could say SPICE IS a sliderule.

[PMA]

Originally posted by john curl
PMA, if you would be so kind to send me a working microcap file on the complementary differential, just whatever state it is in now, perhaps I can load it on my PC and save some time and get comparable results. This would increase the variety of stuff we can try.

Just e-mailed files to you, but I do not intend to make them "public".

[Edmond Stuart]

Originally posted by scott wurcer
Sorry on this one you are incorrect. You're speaking out of context based probably on bad models or poorly implemented convergence criterion. I simply stated that a simple system of simultaneous equations has a unique solution in the general case. Certainly the the five or six node complementary diff-amp when stated as a matix problem with simple square-law and lambda JFETs will have the same answer. Try some time to invert a 6x6 symbolic matrix, you will see why why not just SPICE but MATLAB and MATHCAD exist.

But please elaborate, I learned to solve simple 1-2 transistor circuits with a sliderule. Solving three equations for three unknowns, for instance, pain in the butt sometimes. Typing the same matrix into MATHCAD, I usually trust the answer.

Or I could say SPICE IS a sliderule.

Hi Scott,

Many times I get an error msg. from my simulator: "matrix is singular".
I wonder, have these guys who programmed that stuff, never heard of singular value decomposition (SVD)? Any idea? Are all simulators prone to this (annoying) behavior?

Cheers,
Edmond.

[scott wurcer]

Originally posted by Edmond Stuart


Hi Scott,

Many times I get an error msg. from my simulator: "matrix is singular".
I wonder, have these guys who programmed that stuff, never heard of singular value decomposition (SVD)? Any idea? Are all simulators prone to this (annoying) behavior?

Cheers,
Edmond.

Sometimes too many ideal sources will make a circuit not converge. Our simulator switches modes when it gets this error and tries three or four different integration algorithms and source ramping. I have almost no problems even with some pretty big circuits.

Convergence around the wrong answer is a problem in many fields. When a circuit converges to the wrong answer because I was lazy and put in an ideal current source that happens to be happy to sit at 25000V compliance, I don't blame the simulator.

[Charles Hansen]

Originally posted by Justcallmedad
Charles is this why you switch from folded cascode to current mirrors.

I have to admit that I did not do a critical listening test with only this one change (which is the only meaningful way to do a critical test). There simply isn't enough time to do a critical listening test on every single possible variable in any circuit. My main tool is intuition, along with reports from other designers whose work I respect, although every once in a while I will also do some modeling (SPICE) to test some ideas.

Please let me elaborate on what I call "intuition". Here is one example -- many of my basic circuit concepts are based on ideas that John Curl developed decades ago. When I was a teenager I saw his complementary-differential JFET input circuit published in The Audio Amateur. When I saw this, it immediately was obvious to me that it was not only brilliantly original, but just "felt" right. That is just about the only input circuit I have ever used for my solid-state designs. It has some nice advantages, such as allowing for the elimination of an input coupling capacitor and also the current sources for the differential pairs. It can easily provide balanced output signals (critical for my fully balanced designs) and has a good PSRR. But it is only now, decades later, that in this thread Pavel and John have proven the inherent linearity of such a circuit. (I had good evidence for this in the MX-R amplifiers -- the Stereophile test shows that the distortion for the complete amplifier is down around 0.005% before the output stage leaves class A operation. Using a similar circuit in our new KX-R preamp achieves around 0.002% distortion at similar signal levels. All open loop designs, of course.) The point is that when I "discovered" the circuit and fell in love with it, I was very inexperienced, with very little practical knowledge of circuit design. Yet it has proven to be the best choice for me even many decades later. So I have learned to trust my intuition.

The impetus for changing from the folded cascode is that the current mirrors allowed for a much higher PSRR. When I changed to the current mirrors I also changed many, many other aspects of the circuit. I was happy with the end result, so my conclusion is that the current mirrors could not have caused any significant sonic problems.

Edit: Re-reading this, it makes it sound like I rarely perform listening tests. This is not the case. There are many, many decisions that can only be made on the basis of listening tests. Which capacitors sound best in a given circuit position? Which resistors sound best? Do "snubbers" help the sound? Does a toroidal transformer sound better than an E-I? How does the bias current affect the sound quality? The list is endless. Since proper listening tests can be very time consuming, it is important to narrow their scope. Again, intuition and reports from others can be very helpful here.

[Bob Cordell]

Originally posted by john curl
Bob, I have been using emulation since 1966. In 1971, I took classes at Berkeley from Don Pederson, claimed by the IEEE to be the father of Spice. I just learned from the beginning, its limitations, and never found it as important in making real circuits, although is great for 'what if' questions. I am still not using Spice for anything like a full circuit, and until I get a communication card running on the PC that Nelson Pass gave me (thanks again, Nelson) I don't think that I will be using Spice, except for what PMA has thoughtfully and usefully provided.

John,
You've come a long way.

Originally posted by john curl
Spice emulations? I don't need no stinking Spice emulations! (think bandito icon)
Count me in with Bob Pease on this one. ;-)
I just measure them in REAL-TIME with capacitors and a square wave generator, as I learned to do at my mentor's side, 40 years ago.

[PMA]

Example is better than persuasion.

[john curl]

Now that I have everyone's attention, I'll tell my story about computers.
It all started back in 1963, when I dropped out of school after 3 years of engineering/physics and got a job at Lockheed Aircraft running what, at the time was one of the biggest computers in the world, the IBM 7094. It was a challenging job, as the programs were always acting up and wasted valuable computer time. Once I asked the programmer what his program was doing and he told me wing vibration analysis that required solving a 50th order matrix. I was impressed! After all, when many here were in 'knee-pants' I already had studied linear algebra, somewhat extensively, because of my physics background. I used to solve 3rd order matrices at the bus stop, just for something to pass the time. 4th order was daunting, but 50th order, wow! A lifetime of effort, seemingly. So began my love of computer simulation.
Then, in 1966, many years later, I finished school and was soon snapped up by Friden Calculator to work under 3 experienced engineers, and I would have techs that worked under me.
One, the senior engineer, a grizzled old guy, of maybe 45, carried his slide rule at his hip, like any good engineer at the time. He was the boss, and he was skeptical of the scheme of using the new and powerful ECAP program to validate our circuits. He was used to using his handy slide rule, and that was good enough for him.
Another engineer, perhaps in his early 30's was very excited about computer simulation. After all it was designed by IBM for the US military, and had been extensively used by them already, it must be near perfect! Well I was hired, partially BECAUSE I was so excited about using computer simulation. We were even written up in 'Electronics' magazine in 1966 for being pioneers into the new world of computer simulation!
Well, I became in charge of making the simulations. With the help of the other engineers, and a curve tracer, I made the models for the active devices. It was all DC analysis, so it was easier than it might be in audio, today.
Anyway, I ran the computer myself, after checking that technicians punched cards matched the circuit we were simulating, etc. We were especially interested in what was called 'worst case analysis'. The computer generated a so called 'sensitivity matrix' that showed the sensitivity of a change of any component or voltage, on the final output.
Worst case analysis would plug in the percentage variations that we chose and tried to find the worst case effect on the output. These were discrete digital germanium circuits, and it was a GO/ NO GO situation. The circuit either worked (worst case) or it didn't.
Well guess what? The computer gave us NO GO answers to previously worked out designs, and in 1967, yes 41 years ago, the head engineer whipped out his trusty slide rule and PROVED our results were incorrect! Were we embarrassed, and I thought that we might be fired. So much for trusting computers completely. We double checked our work, of course, but we knew then that if in doubt, for some reason, the computer program would tend to give us a more than worst case result. Probably a good thing.
Today, when I find exclusive use of computer emulation, I remember when I too got into trouble over it. So there, fellow engineers, and quit treating me like an inexperienced kid!

[Justcallmedad]

Originally posted by Charles Hansen
I have to admit that I did not do a critical listening test with only this one change (which is the only meaningful way to do a critical test).

Thank Charles you for details. We all have learned a lot from John and his experience, as well as by you more recently, you show us that it’s possible to get very good results without feedback (I hate feedback). I know that John too, but he must meet THX requirements.

I am doing for learning purposes two preamps that may eventually progress also into an amp, one folded cascode, the other current mirror based, both fet input. I will open a new thread soon, once circuits fully tested and pcbs designed:

[Edmond Stuart]

Originally posted by john curl
[snip]
So there, fellow engineers, and quit treating me like an inexperienced kid!

My dear fellow geezer,

Nobody sees or treats you like an inexperienced kid. Only your view on simulations -in these modern times- is becoming of age.

Cheers,
Another geezer.