[Elecraft] An AGC story

[Kok Chen]

On Dec 9, 2010, at 12/9    9:17 AM, K4IA at aol.com wrote:

> But, since AGC works on every signal in the  passband, if there are  
> two signals - one S9 and the other S3 appearing at the  same time,  
> the S3 signal will disappear because the AGC is knocking down the  
> S9  signal to an S3 level and thereby also reducing the S3 signal to  
> the  noise.

My impression from reading what has transpired is that some people are  
more concerned that when the S9 signal is *not* there, that the S3  
appears to be too loud?

We know that the S9+ signals have to be kept from saturating the A/D  
converter (or even kept in check so the large signal doesn't do  
naughty things to the IF amplifier and the SA612 second mixer -- I  
don't know which of these three is the wimpiest in the chain).  So  
some AGC is neccessary.

Perhaps people will be happier if the AGC response is not as linear  
(in the log-log scale) as the one in the K3 once you have past the AGC  
threshold.  I.e., perhaps people would care less if an S7 signal sound  
as loud as an S9 signal, but they do want the S3 signal to sound less  
loud as the S7 signal.  A non-linear transfer function like that is  
certainly "easy" to implement (as long as I don't have to be the one  
to implement it :-), but will this make more people happy?  Who knows?

Basically, this is what I am simple-mindedly visualizing: currently,  
there are only two parameters to control a piecewise linear curve.

Imagine that there are three parameters, the current two, plus a  
"upper threshold," after which the slope becomes "15"  (i.e. flat at  
S0 to THR, then constant "SLOPE" until "upper threshold" and then flat  
from there until the front end starts to smoke),.  You then smooth out  
the piecewise linear curve using a pair of homotopy between those  
three partitions.  You will end up with (1) a smooth curve instead of  
the piecewise linear one currently, and (2) a way to say "I care more  
about distinguishing an S3 from an S9 signal than I do an S7 and S9  
signal."

Notice that if a simple homotopy is applied to smooth the AGC transfer  
function, and even if "upper threshold" is set to S7, you will still  
hear a difference between S7 and S9 -- just much less than between S3  
and S9.  Also, if the AGC curve is no longer piecewise linear you will  
still hear the S3 signal even if the lower threshold is set to an S3.

I still think the only way to satisfy "everybody" is if everyone would  
homebrew their own rig, or sell an "open" commercial rig where you are  
allowed to make any mods -- hardware, firmware, or software.  See  
first option :-) :-).

73
Chen, W7AY