To: "Paul R. Higgins" higg0008@tc.umn.edu, synth-diy@mailhost.bpa.nl,
From: "Grant Richter" grichter@execpc.com,
cc:
Subject: Re: uA726 replacement, (the comedy of errors)
Date: Sun, 15 Aug 1999 13:57:40 -0500

---

>
> I would also be interested in hearing about people's experiences with
substrate
> heaters (e.g. heated 3046s) vs. tempcos.

Using high grade matched exponential converter transistors eliminates
nearly all of the problems from the current source. Except for that
pesky scale term (3300 ppm) that is left over. The term itself isn't even
that predictable, it can vary a few hundred ppm on either side due to
the transistor doping issues. At first glance, the tempco approach
doesn't look that good. The 3500 ppm resistors are not that close
a match to 3300 and they both vary by +/-100 ppm or so.

So a thermostated pair seem like a good idea. Thermostating eliminates
the term altogether, and should produce an extremely stable current.
A monolithic transistor array seems like the place to start.
Lets try an inexpensive one like the LM3046 which is used in the
Serge NTOs and documented AN-299. Ok we build that and listen
to the oscillator. Hmmm.... there seems to be a distinct "flutter" in
the pitch, the temperature control loop is oscillating and the high
currents are making the supply lines fluctuate by a few millivolts.
Just enough to be annoying.

OK let's overcompensate the temperature control loop, much larger
cap. The oscillation damps out to a slow cycling that just won't
disappear and you can still hear it. Much head scratching occurs
before you realize that because the AN-299 schematic didn't have
pin numbers marked, and you just hooked up whatever, the heater
and sense transistors are all the way across the chip from each
other, producing a large time lag. Move the transistor connections.
Now it works like the Serge. Doh!

This is better and you can calibrate the 1 v/octave setting.
But you discover that the LM3046 is kind of a lousy logging transistor
and you have to overcompensate the VCO at the high end.
So move to a LM394 inside of an oven with an external
heater. Now the thermal lag problem is back and it consumes
a LOT of current, beef up your traces and run independent lines
to the power supply to eliminate interactions. But now the heat
is getting into other parts of the circuit, you put the oven below
the capacitor (Doh!) and the heat causes it to slowly drift.
So get a better cap, ceramic COG works better.

Now come back the next day. When you switch it on, the volts
per octave setting is off quite a bit and it takes about five
minutes for all the thermal masses to reach equilibrium.
And then you can play.

I thought I would post my little comedy of errors to expand on the
fact that designing a heater for something a critical as a logging
transistor pair is not trivial. The thermal coupling between the heater
and sense transistor has to be considered. The power supply lines
to the heater have to be considered. The placement of the heater
relative to other components has to be considered. The LM3046 design does
work as evidenced in Micromoogs and Sources. But those oscillators
are used over a limited range.

By far the easiest design is the tempco resistor. Aside from the cost,
the results are perfectly acceptable. Additionally, perfect stability in
an oscillator is not musical. Acoustic and organic pitch sources
are inherently unstable and the ear seems to expect that. Analog
oscillators exhibit "phase wobbles" in the +/- 1 cent range and
probably account for that "warmer" sound (not the chip heater).