Interesting article on amp distortion by Boyk and Sussman

...about the effects of feedback, both local and global. They are careful to explain the ideality of their models and the limitations they impose. My respect for Boyk has gone up by at least 6 dB :green:

See http://www.its.caltech.edu/~musiclab/feedback-paper-acrobat.pdf

I was led to this one by another one of interest about triode amps available on the audioXpress website:

http://www.audioxpress.com/magsdirx/ax/addenda/muhammad.pdf

I hadn't realized that some articles and reviews that didn't make the cut for audioXpress are on the web. See also a very negative review, Distorted References," of two books out of Berklee.

I speed read the first section of the first PDF

Some questions

What do they mean no feedback every device in there has feedback that you need to think about with the design.

There is the current dependent internal Re of the BJT

The 6dj8 is a frame grid so does it follow the law used in the PDF?

They used simple models if I read the Abstract correctly? you might want different currents for different devices

The use of the 6dj8 and no feedback ??????????

In 1957, Norman Crowhurst [4] observed that since the intrinsic nonlinearity of an amplifier must produce harmonic and intermodulation products from the components of the program material, feedback will combine these products with the program to produce further distortion products.

Click to expand...

And they knew about T.I.M. in the early 50's


cool stuff!

[quote author="Gus"]I speed read the first section of the first PDF

Some questions

What do they mean no feedback every device in there has feedback that you need to think about with the design.

There is the current dependent internal Re of the BJT

The 6dj8 is a frame grid so does it follow the law used in the PDF?

They used simple models if I read the Abstract correctly? you might want different currents for different devices

The use of the 6dj8 and no feedback ??????????[/quote]

I think the point, more or less, in the local feedback sections, is the effect of adding a linear resistance in series with the decidedly nonlinear idealized parts---FET, bipolar, triode. Usually we intuitively think that putting (linear) R in series with devices will linearize them across the board---that is, will improve THD and IM---and this shows that for the given models used at least, that assumption is often faulty, or at best a half-truth.

As far as the models:

The current-dependent Re of the bipolar is concealed in, but present, in the equation listed; bulk resistances etc. are not considered. The FET model assumes that the channel is wholly controlled by the gate, and that there are no velocity saturation effects and other short-channel phenomena. The Langmuir/Childs two-diode-based triode model is a first-order standard and assumes space-charge limiting and fairly orthodox plane-parallel geometries---I think the model is not too far off for most triodes under normal operating conditions, with moderate gm and plate output R, and the DJ8 is pretty much in that category if I'm not mistaken.

I was impressed that the paper was not just a rant about tubes being obviously better to anyone who isn't essentially deaf. He even says something positive about bipolars, which from previous writings I would never have expected.

It's wonderful to connect the residual out of your THD meter to a spectrum display and watch all this in real time while tweaking a circuit.=)

I used to think THD was just a number.

I will have to break down and get a real-time Fourier-based analyzer for this sort of stuff, and find out how close old circuitmaker is to reality. Sussman is no doubt the number-cruncher here, and seems cognizant of the pitfalls on the simulation side of things.

OTOH seeing a lot of this low-level stuff is a challenge even for the mighty Ap gear.

> My respect for Boyk has gone up by at least 6 dB

But see his paper in the bibliography citation 5.

#5 a date of 1899?


For the first section Has the testing been done with real devices? and does the math match the real models?

Yeah I noticed that too and got a chuckle. If I knew the man better I would email him and rib him about it. Reminds me of an ex-friend, retired professor and composer, who had a similar typo in some program notes. The funny part was, the music was so "retro" by then-current academic standards that it was plausible he was a 19th century composer!

[quote author="Gus"]

For the first section Has the testing been done with real devices? and does the math match the real models?[/quote]

What real models? Show me a FET model that is accurate, please.

Particularly the high-performance (low pinchoff high gm low C) parts are not well-modeled by the standard square-law equation.

But---read the paper---they are not claiming anything other than Use these models, ignore other variables, do careful math where practical and careful numerical analysis where the maths are too intractable, and see what the effects are on the spectra. The point is, I think, that we see the IM stuff coming up in a sort of curve-of-growth fashion with signal level, and it is at least plausible that this correlates with psychoacoustics.

A realistic model would include all manner of details, including self-heating shifts in the sand state parts---but it's unlikely IMO that some additional model complexity would drastically change the results. If anything I would expect them to get "worse" in terms of the proliferation of spurious energies.

Boyk is (in)famous, historically, for making his students build super-simple amplifier stages out of tubes and transistors and listen to the results. Of course the tubes always win. That's the context I had when I ran across this, and was pleasantly surprised that it wasn't in the form of a manifesto.

Now clever design, essential with sand state, can linearize things profoundly, especially with well-matched bipolars. I suspect they know that, but that's beyond the modest aims of the already-long paper.

I was impressed that the paper was not just a rant about tubes being obviously better to anyone who isn't essentially deaf. He even says something positive about bipolars

Click to expand...

Careful, that kind of techno-babble might get you moderated here!

> a similar typo in some program notes

Is not a typo. It is a well-crafted fraud. Note the first author and the publishing house.

I typed the wrong thing I meant to type real devices.

FWIW I don't use sims at all its cheating for simple circuits IMO and I don't trust them.

I would think if a paper writer(s) wanted to be taken seriously one would backup the model(s) with some real life measurements.

I just empowered bcarso with some FFT software, so between us, you should begin to see Real World results.

Breadboards are the Spice of Life.


I just can't get my generator to balance on top of that bias battery, like the drawing shows.=)

[quote author="PRR"]> a similar typo in some program notes

Is not a typo. It is a well-crafted fraud. Note the first author and the publishing house.[/quote]

Ah I get it---or as some might say, "No Sh*t, Sherlock!"

Well, o.k.; maybe 4 dB then.

EDIT: Some hint as to the Beekeeper's Press part of the gag, possibly:

http://www3.sympatico.ca/mudthehut/beekeepr.html

I haven't done anything with real parts, but I am able to pretty much duplicate their results in Circuitmaker sims for one of its somewhat more complex FET models (in this case a 2N3823). It will take pretty good-sized signals to get meaningful measurements with a typical sound card, but it is feasible. A decent pair of test signals is needed too.

The choice of their sim frequency ratios, although done to avoid fortuitous cancellations/augmentations, is not too far-fetched in its relevance to music, interestingly, being the just intonation interval of a minor 6th. So if we suppose a ~220 Hz A as the implied fundamental, thus input signal frequencies of 660 Hz and 1100 Hz, the 11th and 13th harmonic energies that supposedly start to emerge with the local feedback R come in at 2.42 kHz and 2.86 kHz. And if audible those would be highly obnoxious to most listeners.

I'm using the sound card input connected the montor output of the Tek THD/IM meter for a better front end, and the abilty to notch the fundimental out as to not clip the FFT. However, I haven't the ability to notch the two-tone signal, so I hope the sound card will resolve ok and not show it's own distortion.

A second, low distortion generator could rock right now. Luckily SpectraLab has custom IM generator freq's for these wacky "Eigertones", but the sound card, isn't the atomic standard for low distortion..

The choice of their sim frequency ratios, although done to avoid fortuitous cancellations/augmentations, is not too far-fetched in its relevance to music, interestingly, being the just intonation interval of a minor 6th.

Click to expand...

Yeah but the inversion of that is a Major-Third, lol. This all means something!

Ok, here are the initial results from an actual real world reproduction of circuits "D" and "F" in the Boyk-Sussman paper. A BJT with a gain of 10, with and without an emitter resistor providing degenerative-feedback. (click to enlarge)

Notice how the addition of local feedback improves the near-in components while creating additional higher-order spectra.

More to follow with fets n tubes!

Lar, did you ever see the Baxandall article?

SpectraLab looks cool. How much does ST get for that?

I have the article now.. processing it as we speak. I think we paid like $500 for this in '98. It's old. I would calibrate it to our primary standard and use it for a console test jig where trained-monkeys (interns) would spec modules after rebuild.

Good FFT for inquiring stuff like this. This is the setup : Here

The moral is that our distortion has distortion on it. It has to.

edit: Baxandall says many of the same things:
"A small amount of feedback (eg6db) in a single ended stage though reducing the second harmonic distortion, and also the total (unweighted) distortion, by about 6db, will increase the higher order distortion, and the quality of reproduction may well become worse as judged subjectively."