Douglas Self on Negative feedback and distortion


I've been reading Douglas Self book on amplifier design and something he said that really makes me think twice.

As you have seen most amplifier makers claim that their amps either does not use global NFB at all or very little of it to improve dynamic (or transient response).

According to Self, the only parameter that matters is distortion and nothing else. I supposed he measures the extra harmonics that the amp produces given a sinusoidal input. In other words, distortion is measured in the frequency domain.

If I remember correctly in my Control Theory course way back in my college days, the frequency domain reponse cannot tell how the amp will response for a given step input. And the STEP RESPONSE is what can tell a lot about the behavior of an amp dynamic and transient response.

In his book, he is very adamant about his position that the only thing that matters is the amp frequency response.

I don't thing frequency response contains information about how any amp would respond to a step input but I could be wrong. Frequency response is only a steady state behavior of the amp. It cannot tell how much the amp would over-shoot, under-shoot, tendency to ringing, and so and so, given a step response. I don't think you can look at the frequency response and make any conclusion about the amp tendency to overshoot, undershoot, ringing and so on...

What do you think?

By the way, I think his book is excellent read into the theory an amplifier design if you can ignore some of his more dogmatic position.
andy2
Self actually says in his book that one should use as little of nfb as possible. He advices that each stage (which in itself has its own natural feedback) should be designed to give as small a distortion as possible. Then after each gain stage has been designed, as a final step, a nfb loop should be added to suit one's needs.

His argument is that global feedback is unfairly criticized by amplifier makers since each gain stage already has its own natural feedback built in. For example, an output stage complementary pair of PNP/NPN transister operating in class A/B has an inherent feedback with its emiter output resister ... So since you cannot avoid feedback in the first place, then why it is so bad?
My takes on this is that global feedback does have by far a larger impact on the behavior of the entire amp as compared to each gain stage feedback.

My one problem (among others) main problem is with his position on distortion measurement.
Hammy,
If you carefully read specs of cheap 100W Yamaha you'll realize that distortion level is measured at 1kHz @1W of output power. Whatever reaches 100W you may only assume what kind of junk you're realy getting. Normally this cheap Yamaha will be in OK shape upto 20W.
I heard a prototype amp based on Slone's OPTI-MOS amplifier circuitry in my system. It was class AB with some level of NFB. It was the most transparent amp with superb Bass control and decent treble extension that I have heard in my system. It did not add anything and did not take anything away from the sound. I'm waiting for the production version to emerge. Watch www.zusaudio.com for details - they are just starting up.
Eldartford wrote:
"Frequency response is not the same thing as step function response, but it is correlated. And a lot easier to measure. If an amp is 0.5 dB down at 100 Kc it probably has good step response."

You cannot make this inference. Step response and impulse response are dependent on short-term transient power delivery. I can design you an amp that has bandwidth to 1 MHz, but has lousy step reponse (overdamped). To achieve critical damping, the decoupling must be done right (topology), and using components with optimum characteristics for that. This is what is wrong with 99% of consumer gear. The designers understand circuit design, but not transient power delivery.
It's been a while since college. But I'll put in my 2 cents here.

For a close-loop amplifier design, one is mostly concerned with GBW (gain-bandwith product), which is a constant for an amp. You wouldn't run an amp in open-loop where the gain is max and bandwidth is min (half-power bandwidth). You extend the bandwidth by reducing the gain (dB) with measured amount of feedback (i.e. close-loop). Of course, bandwidth enhancement may not be an issue for audio frquency range.

Remember that there is always some parasitic cap in solid state devices that causes Miller-Feedback which is unavoidable. As for local negetive feedback in a multi-stage amp, it is usually employed for, amongst other reasons, attaining certin amount of gain per stage as well as slew-rate control. How the step response will look like can be predicted by the slew rate of the amp, or by how fast the amp is.

Global feedback, which I guess is taken from the very last stage, can be seen as error-correction from a Control System point of view. It is also a way of reducing the effective output impedence of the amp.