Distortion spectrum vs feedback

[artosalo]

Originally Posted by DF96

Bear in mind that -6dB is far too little feedback. You either need none, or enough. With none, you just get the valve non-linearity (mainly low order). With enough (-20dB?) you get significant distortion reduction. At -6dB you get lots of high order products arising from the feedback, but little suppression of them. Worst of all worlds.

I saw this comment at an other tread. This is not the only I have seen claiming that low amount of (global ?) NFB will increase the level of higher order harmonics.

One documented article can be found here: http://www.passdiy.com/pdf/distortion_feedback.pdf

There is results (at page 10) about an experiment done with class A SE Mosfet amplifier showing the increase of higher order harmonics with low amount of NFB.

I made my own test with a 15 W UL-connected tube amplifier to see if I can see this phenomenon.

I made the test with constant -3 dB (7,5 W) output level. Global NFB was adjusted from 0 to 22 dB and the level of harmonics up to 7th was read on each NFB level.

The result is that my experiment did not show such behaviour. Instead the level of all harmonics decreased when the NFB was increased.

The results are here:

An externally hosted image should be here but it no longer works. Please upload images instead of linking to them to prevent this.

http://kuva.termiitti.com/image/17762.gif

I also have the spectrum plots at each NFB if somebody want to see.

Why this difference ?

[TheGimp]

It is going to vary based on the harmonics generated by the device without feedback.

Feedback modifies the terms, so it will vary depending on the device used.

[Yvesm]

Originally Posted by TheGimp

It is going to vary based on the harmonics generated by the device without feedback.
. . .

. . . which are already amazingly low . . .

[the_manta]

Thats exactly what I observed, too
Even little Feedback decreases the overall distortion at such a rate, the little bit, which could be added is insignificant. Thus, IMHO Feedback is and has always been essential for a good power amplifier.

[Bob Richards]

Using just a 1kHZ sinewave as a test signal may not show a few things. At supersonic frequencies where the feedback is likely to roll off, the distortion spectrum may get ugly. Now that so many program sources are digital, imperfectly turned back into analog, and may contain a small amount of this supersonic or even Rf energy, it could be important to look at. Also there's intermodulation distortion that gets generated by any negative feedback "corrector" circuit (usually the front end). But the person who found that 6dB of neg feedback is worse than 0 or 20dB may have done a sloppy job of setting up the measurement process. Not everybody realizes all the variables at play.

Personally I'd rather have as much as 0.2% harmonic distortion than to have high I.M., any chance of slewing related distortions exasperated by high feedback, any chance of overload related problems or phase margin issues... My own tests showed higher order distortion product went up when neg feedback was increased. With no feedback only the 2nd and 3rd harmonics were above the noise floor of a $30,000 HP spectrum analyzer. I'm sure this varies with topology and a few other things.

With tubes, the only place I ever use neg feedback is in the output stage when it's a push-pull topology, and I use as small a loop as is practical. Part to minimize crossover distortion, and part because the output stage is driving the substantially reactive load of the speaker. It may be better to have a lower output impedance for driving such a reactive load, or not, depending on all the variables.

[gaetan8888]

Hello

There is this graphic about nfb vs thd.

Bye

Gaetan

[Bigun]

wow, a very interesting measurement. It reminds me that I read somewhere else another person had looked at a solid state Class AB output ad also found that any amount of gnf was good and the more of it the better. In other words, there are real life situations where measurements don't match the theory of simple topologies.

I also notice that despite Nelson's article on gnf he consistently uses gnf in designs that I've seen (e.g. Zen, F5) so again, there's more to gnf than theory.

[SY]

A theory is valid as long as its axioms are understood. The "increasing order" stuff is true for pure square law devices which tubes are very much not.

[gaetan8888]

Hello

I usually do my transistors amps with an open loop of 60-70 db and a feedback of around 30 db. Tube amps can be done with none or very low feedback.

Bye

Gaetan

[artosalo]

I already got some questions as Private Message about the amplifier used at the tests.

Here is the schematic:




Kuva 17764 - Kuva.termiitti.com

And one photo, (where the opt is Edcor, not the Hammond used at the tests).

An externally hosted image should be here but it no longer works. Please upload images instead of linking to them to prevent this.

The reason for low distortion values - even without GNFB - is that all stages were optimized before any tests were done. At the voltage amplifying stage (6J8) the screen voltage is adjusted to the "sweet spot", similarly the bias and the bias balance of output tubes were optimized. The optimimum total current of 6P9 was 20 mA measured at the cathode.