More on Fav Designs

[Paul Schreiber]

Hot Damn! Some intelligent activity in a group (there is hope).

VCOs
====

First, no VCO has a "sound", sterile or otherwise. They may have THD, but a pulse is a square is a saw. But what they DO have is drift and have linearity errors.

Now even the CEM3340 will drift some. But not as much as most, because there is a circuit on the die that generates the +3300ppm voltage. Note that this must be MULTIPLIED with the input voltage. A resistor divider is a multiplier (by a coefficient <=.9999). The on-chip multiplier (OK there goes my trivia question: where did OnChip get their name) has linearity errors. This is also called a "tracking error": instead of 1V/oct, get .993 in the first octave, .997 in the second, 1.002 in the third, etc.

I think any synth "voice" needs 3 oscillators. Monosynths with 1 or 2 just aren't very 'fat'. Polysynths using CEM3394s are 'thin' in omni mode because a 3394 only has 1 VCO per voice.

Gene's point is for small # of VCOs, tracking can suffer some, and I agree 100%. Now what remains is the nitty-gritty.

This is a QUICK overview.

a) #1 overlooked component in a VCO is the integration capacitor. You just can't chunk any old cap in there. Why? The quick answer is that real-world caps have ESR (effective Series Resistance), leakage, temperature drift (just like a transistor!) and (what we worry about) something called dielectric absorbtion. All of these contribute errors.

The best cap to use is a metallized polypropolyne.

b)#2 is the integrator used in the op-amp. A perfect opamp has zero input bias current. But in the real world, the input bias current "steals" current into the integration cap. So what? Well, the problem is this input bias current DOUBLES every 10C. And, for low frequencies, the input bias current on cheesy opamps can approach the integration current.

c)#3 discharge transistor. Best advice is use a DMOS FET, like an 2n7000. They have a factor of 10 (to 1000!) lower ON resistance. This eliminates errors at higher frequencies.

The 3080/LM13600
==============

Hey, these are OK for poking around. Heck, I bought them in tubes of 100 for years. But they suffer from:

a) control voltage feedthrough. This translates to audible 'thumps' when the CV rapidly changes (ie fast attack)
b) they like current SOURCES to control them, which are harder to manage (but doable)
c) their THD/noise is between AM and FM radios which is OK if you can live with it

I like quiet VCAs with NO feedthrough.

The 3310 EG
=========

All agree, this is the chip. So concern over the availability. Gene, I GOT 1200!! Now, this is certainly enough for now. So don't be shy using it. I will design an equivalent from discrete later this year.

Power Supplies
===========

It's hard to beat a 723/D40xxxx pass transistor. This is a case where 'vintage' is better. I will ask Walt Jung for permission to publish his "super high tech" power regulators. They have 1/1000th of the residual noise of a LM7812. Amazing!! Can be built for about $8/side.

Running 500 Amps to heat a transistor
============================

Granted, this is the quick way to do it for 1-2 VCOs. If your building 2 VCOs, be sure the both exponential gen transistors & heater are all on the same die. In order for this to work, the DIE temperature needs to be about 60C which will burn your little pinkie.

Doing this is a polysynth is, what is referred to in engineering circles, a bad idea.

Keep the dialog rolling.

Paul Schreiber