On Sun, 18 May 2008 22:41:24 -0500, flipper <flipper@[EMAIL PROTECTED]
> wrote:
......snip!......
>
>I don't know how you got the idea that local NFB has no effect. Local
>FB works just like global with the difference being how long the loop
>is.
>
>Your basic gain vs distortion math is correct but it applies to any
>FB, not just global. I.E. Apply local NFB around (or in) one stage
>and the distortion of that stage is reduced in the same manner as
>overall amp distortion is reduced by global NFB (or vice versa with
>PFB). e.g. Take just a single triode. What would the gain be without
>the Rk (using, say, fixed bias instead)? Now, what is it with the Rk?
>That's how much NFB the Rk adds and distortion is reduced by that
>amount. Or, conversely, that's how much increase in distortion there
>is when you bypass Rk to get more gain on that stage.
>
>However, it doesn't do you any good to have a squeaky clean front end
>and then feed it into a heavily (by comparison) distorted power stage
>because distortion over multiple stages is the square root of the sum
>of the individual squares so if one section is significantly higher
>than the other it dominates the overall distortion figure (and by more
>than if it were a linear relation****p).
>
>But you also have to have the gain in order to apply global NFB so
>it's better to use some local PFB, to get the gain with some increase
>in distortion of the squeaky clean stage and use that acquired gain to
>apply global, thereby reducing the overall distortion.
>
>The fallacy some engineers fall into is observing that equal PFB and
>NFB cancels so they conclude you gain (pun) nothing. I.E. You add
>10dB, then subtract 10dB, and you're right back where you started. Or,
>when not equal, why add 10dB and subtract 18dB when just subtracting
>8dB gets the same result? And that would be true if it was the same
>loop but what they miss is the PFB is local, over the much better
>stage, while the NFB is global and, as mentioned, the combined
>distortion is not linear but the square root of the sum of the
>squares.
>
>Of course, there are other ways to achieve the same result but I get a
>kick out of this one because, component wise, it's 'free' as the
>difference (besides picking good component values) is simply
>terminating the phase splitter into Rk rather than ground.
Can I summarize this as follows? :
A linear disturbance (no sqrt of sums of squares) would not benefit
from juggling gains of local feedback paths within a global feedback.
This assumes that you haven't messed local gains up so badly that
overloading or slew rate limiting occur.
A disturbance that adds non-linearly (square root of sums of
sqyares) WOULD benefit. That makes sense, you would want to linearize
(more local feedback) the offending section, and boost gain (lower
negative feedback, or POSITIVE feedback) in the other amplifier
sections to make up the required overall global feedback or loop gain.
Now here's MY problem..... it's a mathematical one. It's my belief
that square root of sums of squares applies to uncorreleated signals
like white or thermal noise. Distortion is correlated (it's a
ploynomial base), and I don't think that the non-linear sqr root of
sums of squares (RMS) applies here.
For noise issues, yep, I have no doubt that juggling gains will
help, but for non-linearity, it's still my belief that loop gain
(overall) is what will reduce your distortion.
This is a tricky question, and it gets down to some hairy
theoretical issues..... do you know for sure that non-linearity
(distortion) is in fact added up in the sqrt or sums of squares? You
can decompose nonlinearity into a fourier series, and treat the
distortion as added sine waves, and they aren't random!
It was nice that you clearly backed up your assumptions, so I can
argue the point without getting personal !
Thanks,
Paul
|
