AW: [sdiy] Dual Frequency Shifter update

[Magnus Danielson]

From: jhaible at debitel.net
Subject: AW: [sdiy] Dual Frequency Shifter update
Date: 19 Jul 2006 11:05:12 +0200,Wed, 19 Jul 2006 11:05:12 +0200
Message-ID: <1153299912.44bdf5c8a0ae0 at www.debitel.net>

> > The Raytheon RC4200 used the Log-Antilog scheme for multiplying as  
> > the sum of logs. That is x * y = antilog((log x) + (log y)) 
> > They have  
> > to be DC biased for single quadrant. It had the lowest rated  
> > distortion (0.1%) of the commercial multipliers (vs. 1% for AD633).
> 
> The circuit that makes a 4Q multiplier from a 4200 is quite complicated.
> You need a lot of matched resistors, and a lot of trimming.
> >From memory (I played around with a 4200 some years ago), you also
> need to impedance-match the offset trim inputs to the signal inputs
> because of the bipolar technolgy (bias currents).
> 0.1% distortion looks good, but it doesn't say anything about carrier
> bleedthru, and the quadratic term of carrier bleedthru is a critical
> parameter in clean audio 4Q multiplication. The AD633 is quite good
> at this. Mind you, I don't know if the 4200 is worse (or better) than
> a 633 in this - I just say as the 4200 isn't a 4Q mult per se, this 
> parameter can hardly be specified for the _chip_; only for the complete
> circuit.
> Also, you must avoid running out of the 4200 allowed input range
> by all means. When you overdrive a balanced modulator, it just clips.
> When you overdrive a 4200, it oscillates. So you need extra headroom,
> which will take away from your system performance.
> 
> BTW, the 4200 is used for the Ring Mod in the Korg MS-50.
> Seems to work well there. But the demands are certainly lower than
> for a FS application.
> 
> 
> > What if you used a single log conversion BEFORE the dome filters,  
> > then summers, then anti-log, then sum and differential 
> > addition. Dome  
> > filters should be DC linear, so logging before won't matter? 
> 
> I fear Magnus is right with what he replied.
> I've made simulations to see the effects of rectifier ripple in
> a companded FS system, and the problem is that unlike in ordinary
> companded channels, the distortion from the compressor is not
> partially cancelled in the expander, allowing short time constants.
> 
> Using nonlinear functions instead of a compander is equivalent to
> a compander with zero time constants, I think.
> So even though I cannot see how it could possibly work, I still
> have a faint hope that there is some ingenious trick where everything
> falls into place.
> Tim ?!

Even if you have a nice non-linearity which is bipolar (like direct compression
without timeconstant) the distorsion it causes will probably be similar to
what you have simulated. You can only compand a small amount with such a system
before the distorsion exceeds the S/N gains that you hoped to acheive. You are
in effect going into the non-linear distorsion land this way. There are cases
where the non-linear companding work, but they really don't involve filters in
signal frequency range, because those will break the assumptions that makes it
work with any fidelity.

I am highly sceptic about this route.

The direct companding that I mentioned above is this: precission rectifier,
companding curve (linear/non-linear but no filter) and VCA. I think it is the
closest you can get, but not near what you want.

Sorry for spoiling your party. I'll bring some beer next time, OK?

Cheers,
Magnus