[sdiy] discrete SSM2018?

[jhaible at t-online.de]

> Hello feedback haters, here's an idea to flame:

He he, this could be me! (;->)


> Therefore the idea is to use two gain cores and to feed 
> them 
> with dependend tail currents: a*I and (1-a)*I {a: 0...1}
> so that the sum is always constant I.
> If we now add up the signals of both gain cores, this sum 
> signal 
> will also be constant. Now we are able to create a 
> feedback loop. 

I haven't thought of *that*, but - strange coincidence -
I have been thinking about a similar topology lately, which
is also used in some of the better spec'ed VCAs:
Use two variable gain cells in parallel (inputs connected),
and run one with a fixed gain and feedback, while the other
one has variable gain and no feedback - this is the real
VCA part. If both gain cells are identical, the distortion
would be reduced in the VCA to the same amount as the
"reference path" with feedback. It's another form of 
pre-distortion, but youcan drive the inputs even harder than 
with simple diode pre-distortion, so a better SNR is to
be expected. I think I've seen this in another SSM chip
(one of the "modern" AD ones, too.)

I think the two methods are closely related. Maybe feedback
from both channels (a and 1-a) has advantages, because
both cells are inside the linearisation look ?
But then again, slight differences between the two gain
cells will shoot the performance down in either option,
won't they ?

> This could be translated into using two of the 13700 otas 
> in parallel 

I have seen such a circuit in a magazine some years ago,
with 13700, no kidding. The article was simplifying
a lot of things and didn't have an analysis of side effects
from mismatch. So I was fast to peg it as another circuit
that looks nice in theory but dosen't work in practice.
Maybe I was too fast in dicarding it.
Anyway, I've recently designed a circuit with a pair of
differential pairs to act as two gain cells, and just
a few transistors insteads of opamps for the feedback part.
(This follows the unity gain - variable gain approach, not
the a, 1-a approach.) This looks very promising in simulation,
and it "only" needs a tightly matched quad transistor array,
but it's too soon to be more specific: No breadboard test, and
not even simulation of slightly mismatched transistors.
But I think it's an interesting concept, and as you mentioned it,
I want to encourage you.

However, there is one little detail left (to quote you again):

> Hello feedback haters, here's an idea to flame

In fact, I have heard complaints about the sound of such,
or similar, VCAs from people who use them and don't necessarily
know about the feedback issue (which otherwise might cause some 
preoccupation), so we *might* (and I *do* have some preoccupation
here!) get a VCA with excellent specs in the end, which has
an unpleasant sound quality. But note the "might" - it's surely
an interesting idea to follow.

JH.