Feedback delay & distortion

[Jorge]

One doubt many people have is if feedback isn't instantaneous, then how can it reduce distortions?
(BTW, feedback IS NOT instantaneous if there's some doubt about it).

To show feedback in action, I've simulated the simple circuit below, with parameters taken from the well known Leach amplifier circuit (the amps' BW is 8 MHz).

[Jorge]

The figure below shows input signal (sin) in green, feedback (fbin) in red.
It's a 10kHz square wave, band limited to about 40kHz.

Just by looking at the figure, delay becomes aparent.

One can see there's a delay during the transition time (rise and fall) and it disapears when the signal is stable (flat top and bottom).

[Jorge]

Here's the error signal (sin-fbin) applied to the amp to correct the error between input and output.

The error signal reaches 160mV at the beggining of the transition in less than 1 microsecond.
It is this large error signal that forces the amp to 'catch up' with the input, from there on the amplitude decreases until it falls to zero when the transistion has ended (flat top or bottom).

An amp with a wider BW will be faster (less time to catch up) and one with insuficient slew rate may distort because the error signal cannot rise fast enoug.

Are there negative aspects to this one microsecond delay?

I believe not, witness the many highly regarded feedback amps used all over...

[sam9]

It's an obvious concern. I console my worries by the following train of thought.

Q1- What is the physical distance the feedback signal must travel?
A1- Well, in my board layouts it tends to be between 10mm and 50mm.

Q2- How fast does the feedback signal travel?
A2- I really don't know but it most certainly is some significant fraction of C.

Q3- For starters let's assume that the initial signal suffers no distortion. A delayed feedback signal will actually creat distortion. How significant is this?
A3- Well, suppose the feedback signal traves at 0.1C (probably way to small), or 30,000,000m/s. It will take 0.05/30,000,000 seconds to meet up with the original signal, that's 1.7 ns if I kept track of all the zeroes. It takes 50,000 ns to complete a full 360deg cycle of a 20kHz signal. 1.7/50,000*100= .003%. This is NOT a distortion figure; that would be much smaller.

Anyway, I'll relax and go back to sleep now.

[Frank Berry]

Feedback cannot be 100% effective because it takes time for the audio signal to go through the amplifier.

Fortunately .... for audio frequencies, the delay through the amplifier is short enough to make inverse feedback work pretty well.

[sam9]

Going back to my calculations, if they are approximately correct the difference between the actual delay and zero delay is so small the delay can be considered "virtually" non-existant. (For the microwave guys I can concieve that it is another matter entirely)

I really like the word "virtual" as in "virtual reality", or "virtually dead" (ref. M Python). I'm not quite sure what it really means but seems to be a servicable way of removing some of the knats in life's oinment.

[MBK]

That's very interesting - I have never seen the actual time involved calculated or simulated. The interesting part is the error correction signal and the time scale. 1 usec is actually very slow - how can that be for a reasonably "fast" amplifier? The shape is horrendous too, I can imagine all sorts of bad things happening here...

1 usec is within the range of say, offset delay errors for speaker drivers (e.g. a tweeter offset of 5 mm to the woofer will have a delay of 1E6*0.005 m / 343 m = 14.6 usec. I would have thought electronics a lot faster than that.

BTW for all I know the delay is not related to the electron travelling time and speed and distance... It's the finite "reaction time" of the active devices, and reactive elements, creating group delay and phase shifts.

[tschrama]

Will this bullsh%t about time delay ever stop?

Consider this:

Even in tube circuits, when electron flow at a relative inmensely slow speed through the tube semi vacuum from cathode to anode, still the time delays is of no concern for audio!

just because some japanse (chinese?) guys makes money of a 20$ chip amp that he sell for 2000$, doesn't mean that he rediculus theory about time delay and feedback is even remotely connect to any thrueth



PS


Q1: does a spice simulator actually take in to account the speed of electrical signals?...

Q2no? How can your simulatro then give any usefull info in this?

[pinkmouse]

Speed of propagation of a field/signal in copper wire is C. The actual electrons however, only move at about 1cm per sec...

[andy_c]

Originally posted by tschrama

Q1: does a spice simulator actually take in to account the speed of electrical signals?...

Q2no? How can your simulatro then give any usefull info in this?

If macro modeling of the feedback amp were used, it's possible to use the SPICE transmission line model somewhere inside the macro model. If the line were matched at both ends, that would simulate a constant delay independent of frequency.