Negative feedback (NFB) - that's a big can of worms. After all, every audiophile knows feedback is evil, and you must have no global feedback.
Which is of course nonsense. All you are doing is reducing global feedback with more local feedback. So how did the low NFB being good myth come about?
Well firstly we have to be careful - some people like the sound of distortion - particularly if you are creating oodles of 2nd. So a report of NFB being bad may be down to a preference to higher distortion. But lets ignore that particular issue, and run through a thought experiment.
Now let us imagine you want to design an amplifier to drive a headphone or loudspeaker. And to eliminate the possibility of people liking distortion, lets assume that you have a target of 0.01% THD, that you design two amps, one low NFB, one high NFB, but both must measure with about 0.01%.
So you have a high NFB (normal) design, and to make it low NFB you use more local feedback - say in the case of bipolar use more emitter degeneration. And because there is less NFB, you can increase the open loop frequency response. But you find it will not meet the 0.01% THD, as the major source of distortion is the OP stage and there is not enough NFB to correct the OP stage - so you improve that open loop performance by either more local feedback around the OP stage, or increasing the bias current.
So now we have two amps, both measuring about the same, and I can say that for sure the low NFB amp will sound a lot smoother and more refined. So that means that low NFB must be better, surely?
No it doesn't.
Let's look at what you have done to make it low feedback.
1. Local feedback by using emitter degeneration. This makes it more linear, but also more linear at RF frequencies - and that will reduce noise floor modulation - and that will make it sound smoother.
2. Increasing bias current in the OP stage - this will reduce the HF distortion, as the annoying thing about crossover distortion is it sounds very hard due the very high frequency harmonics - indeed, Class B operation has distortion extending to infinite frequencies. So increasing bias will also make it smoother.
3. Increasing the open loop FR - this will mean that there is now more feedback available at high frequencies - and this too will make it smoother as HF distortion will be lower.
Now the issue of noise floor modulation I have talked about a lot. But high frequency distortion is also very important - indeed I always measure up to 2 MHz looking for distortion, as distortion at very high frequencies is audible. By that I mean whenever I have reduced distortion at say 100 kHz it sounds smoother. Which is crazy - how can distortion at 100 kHz be audible? Well it depends upon how that distortion is there, and some very high frequency distortion acts via changing the delay - so the delay then changes with signal level, and you can only measure this by looking way out of band. But the delay change with signal is highly audible, (it again sounds smoother by eliminating it) and its audible because it is modulating the timing of transients - not because you can hear distortion at 100 kHz.
So NFB is categorically not a problem at all, but like most things in audio, is actually very complex. People would be better off talking about minimising distortion, noise floor modulation, and HF and RF distortion. But that doesn't make for a snappy line on a sales sheet.
Rob
