The belief that zero feedback amplifiers are in some way sonically superior to those with feedback is often debated, but looking at some 'zero feedback' designs it is clear that feedback still exists, often as local feedback applied to single stages. Simply adding an emitter degeneration resistor to a bjt common-emitter stage adds local negative feedback, and even if no resistor is added there will always be a small internal resistance in any real transistor, so feedback always exists (unless we add some sort of 'negative resistance' circuit to cancel the effect). The question then remains whether local feedback avoids any of the problems claimed to be important for global feedback. The addition of high order harmonics is one of the most common criticisms of feedback, so this is a good starting point.
I started doing a few calculations to demonstrate that adding a source resistor to a single fet common source stage would add high order harmonics in the same way that adding overall feedback to any amplifier with a square-law transfer function adds these harmonics. I got as far as showing that the original sinewave had an infinite series of higher order terms added, but fortunately avoided the need for some more difficult calculations when I found an article Small-Signal Distortion in Feedback Amplifiers for Audio by James Boyk and Gerald Jay Sussman, (a pdf file), which shows the results of numerical methods used to produce intermodulation distortion spectra for feedback amplifiers. They were not analysing local feedback in order to compare it to global feedback, they just happen to use devices with and without degeneration resistors as one example to demonstrate the effects of negative feedback. They mention that they use idealised device models, so their conclusions fail to include the high order harmonics already present in a real 'not exactly square-law' fet. The fet analysis includes only two levels of feedback, 1.8dB and 9.5dB, so there is no indication of how their conclusions relate to high feedback levels, but at these low levels a range of higher order intermodulation distortion products are added. One important conclusion is that adding small amounts of negative feedback to a highly non-linear amplifier can be a bad idea, and the question of whether local feedback avoids the problems of global feedback is also partly answered by this article, and with regard to high order harmonic generation the answer appears to be 'no'.
It was pointed out by Baxandall many years ago that although high order harmonics are added as low levels of feedback are applied the amplifier stage becomes more linear. This sounds like a contradiction, but the original curved transfer function straightens out and becomes closer to a straight line in a smooth and continuous way as feedback is increased, whether local or global, even at low feedback levels where high order harmonics are being added or increased. 'More linear' does not necessarily mean less audible distortion, and increasing high order harmonics are not necessarily an indication that we have added sharp discontinuities in the transfer function. An exception is if the original open-loop nonlinearity is so high that it seriously reduces feedback loop gain over part of the signal voltage range, an example being clipping, and then increasing feedback can increase high order distortion and also sharpen the clipping.
I mentioned in my articles on symmetry and input stage distortion that although a single ideal fet produces only second harmonic distortion, if we use two in a differential input of the 'long-tail pair' type then if they are accurately matched the second harmonic can be cancelled, but then third and higher harmonics are added. The 'Small Signal Distortion' article mentioned above includes another revealing result, for an example of a differential fet input stage the intermodulation distortion spectrum has a wide range of high order products, missing some even order components but extending beyond those added by the previous example of a single fet with feedback. My conclusion that this is not the best choice of input stage appears to be supported by this result. Using high gm types and including source degeneration resistors should give some improvement in distortion originating in the input stage, as should a high gain-bandwidth product for the overall feedback loop so that the input stage differential signal is reduced. It is generally found that high order distortion reduces faster than low order as signal level is reduced, so high gain-bandwidth amplifiers can be a great benefit for reducing input stage distortion.
How harmonic distortion changes at higher feedback levels has already been covered, including results published by Baxandall, the result for his example being that after an initial increase in higher order distortion it eventually starts to reduce as feedback is increased, and at over 80dB feedback all the high order components remain far under -120dB. Curiously the Boyk and Sussman article states that even the 9.5dB feedback results are lower than at 1.8dB, but without showing the spectrum at the higher level. Looking at their figures it appears that they use a fixed input voltage and increase the output load impedance to keep voltage gain constant at higher feedback, while a constant load plus increased input to give the same output could give results closer to the Baxandall figures.
The suggestion in the article that the high order components can add a subjective effect something like a modulated noise level may be true at low feedback levels, but my own observation of listening to the distortion extracted from my high feedback mosfet designs was that the noise heard was just a continuous unchanging level with no sign of anything related to the music signal, even though the test method rejects much of the amplifier noise, and so any distortion effect must be well below the thermal noise level. In this case a high feedback level proves to be very effective in reducing all forms of distortion far below audibility.
The idea that high order harmonics sound far more unpleasant than lower order appears to overlook an important distinction. Just adding a source resistor to a fet, as mentioned above, actually straightens the transfer function and makes it more linear, and yet adds high harmonics. An alternative way high harmonics can be added is by a discontinuity, as for example in crossover distortion. I am not aware of any comparative listening tests to determine whether one cause of high harmonics is more unpleasant than the other when the peak distortion levels are otherwise similar. The way the distortion varies with signal level will certainly differ, the higher harmonics falling fastest as signal level is reduced for the smoother transfer function, while for a discontinuity the high harmonics will not fall to the same extent, if at all. When music has low average sound levels with occasional high peaks most of our listening is done at low levels, so it seems probable that crossover distortion in class B and to a lesser extent in class-AB amplifiers is a far more serious problem than the high order effects of low level negative feedback.HOME.