AW: [sdiy] Dual Frequency Shifter update

[René Schmitz]

Hi Jürgen and Magnus,

jhaible at debitel.net wrote:
> This will work?
> Then it's probably _the_ solution! 
> (No problem to build a matching triple (2 before mults, 1 after sum of mults)
> of nonlinearities from a couple of transistor arrays, I guess. First thing
> that comes to mind is just two anti-parallel diodes in the feedback of an
> opamp. 3 x 2 diodes, made from 2 MAT-04s.)
> And it really will work, conceptually??

I doubt that. A nonlinearity has the property of being nonlinear. :-P 
I.e. in general f(a*x+b*y)!= a*(f(x))+b*(f(y)). (Try it with x^3, exp ...)

The multiplication stretches and squashes your predistorted signal 
across the reconstructing inverse nonlinearity. Only at unity gain will 
you have your original signal back. The adding of a second signal alone 
will prevent this. So you need a=1, b=0 in the above if you will.

On the predistortion nonlinearity you'll get intermodulation products 
for any non-sinusoid tone signal, some even with higher orders, landing 
between harmonics, these would be modulated in the multipliers 
generating sidebands, regardless if they've been dome filtered or not. 
Its not the dome filter who moves them around, but the multiplication. 
(Without the dome filter they'll likely generate two sidebands, instead 
of one.) The same thing will happen again in the reconstruction 
nonlinearity, just you have way more partials now, some of them moving 
around...

If it was possible to cancel the predistortion, then you could use the 
clipping of the multipliers as your predistortion, and cancel it at the 
output. :-)

>>The dome filter is probably a contributor to bad S/N so you really want to
>>replace that 
> 
> 
> I don't think it contributes any noise worth mentioning.
> The signal runs thru 6 opamp stages with a max. noise gain of
> 2 in each stage. I don't hear any noise even with cheap TL072s.
> If noise is an issue, I can still choose resistors in the 600R
> range and go for low noise opamps.

Why do you need to compand at all? (multipliers?)

Maybe you could have a multiband companding, to get good response times 
while preventing distortions. (Two bands should be enough.)

Cheers,
  René

-- 
uzs159 at uni-bonn.de
http://www.uni-bonn.de/~uzs159