[sdiy] modular synth "standards" (longwinded)

[Grant Richter]

(I've always meant to write this down anyway, and it's -20 F outside,
so long winded mode on)

Here is a compilation of hearsay from I got from a bunch of different
people(1) It's technical gossip with no "official" standing.

Music synthesizers "evolved" from the analogue computer. Musical waveforms
are basically "calculated" by analogue methods (shortened to analog in the
US). Original tube analogue computers used 0 to +/- 100 volts for their
measurement range. This fit nicely into base 10 mathematics, and 1 volt =
1%. 

When solid state operational amplifiers arrived, they worked at lower
voltages. One problem is that transistor op-amps don't work as well near the
supply rails. They needed some voltage "overhead" to work most linearly. So
+/- 15 volts was selected as the op-amp supply voltage, which allowed use of
the op-amps over a +/- 10 volt range with 5 volts of overhead, and still fit
nicely into base 10 mathematics.

Analog computers use "coefficient potentiometers" to set computation
constants. In solid state units, these are 1000 ohm which draw 1 milliamp
per volt and is reasonably low power. The input resistors on the op-amps are
normally 1 meg ohm. So that the loading error of the input on the
potentiometer is 0.1% 1K loaded by 1 meg ohm forms a voltage divider with a
0.1% error.

Analog synthesizers also use 1K resistors like analogue computers. But since
accuracy is not as important as noise, the 1 meg input was probably lowered
to 100K ohms to help reduce noise pickup from unconnected inputs. Many
synthesizer modules have 1K resistors in series with the outputs to limit
short circuit current when the patchcord tips contact the grounded faceplate
or a connector ring (If the output is at 15 volts and the tip is shorted to
ground, the 1K resistor disappates 225 milliwatts. 1K is the lowest 1/4 watt
common resistor value that will not fail under a continuos worst case fault)
Also 1K output resistances and 100K input resistances attenuate by 1% at
each connection. That means you can have 50 modules in series before your
signal amplitude drops by half. (The 1K resistors on outputs also allow two
output signals to be mixed just by shorting them together with a multiple
jack)

The range of human hearing is usually agreed to be 20 Hz to 20 Khz which is
10 octaves(2). When it came time to pick a standard, 1 volt per octave was a
no brainer because they were already using 10 volts as a standard. So a 10
volt control signal will sweep a 1 volt per octave oscillator over the whole
frequency range of human hearing. An octave (2:1) expressed as a voltage
ratio, is 6 dB, so a 10 volt control signal will change the attenuation of a
logarithmic voltage controlled amplifier over a 60 db range. A good fit for
the dynamics of human hearing also. (some systems use 12 dB per volt to
allow 5 volt envelopes to cover 60 dB)

Early modular synthesizers did not have all the desired functions available
as modules. The early buyers were concerned with being able to interface
directly from any module to external equipment which used 1 volt peak to
peak as a standard. So purchasers specified that the equipment should have 1
volt peak to peak audio levels. That way it could be directly connected to
their existing equipment. This is why the early Moog and Buchla systems use
these levels for audio signals. (The idea of using audio and control signals
interchangeably happened later)

It is expensive to make analog electronics with a signal to noise ratio
greater than -60 db or so. A 10 volt peak to peak signal is close to +20 db,
so if you passively attenuate it by 20 db (back to 0 db at the mixing desk)
your noise floor is also attenuated. This gives you a "free" signal to noise
ratio improvement of 20 dB, so running signals "hot" is a good idea. It also
allows you to use audio and control signal interchangeably. (Another added
bonus is that a pair of headphones can be plugged directly into the module
outputs to check them, the 1K output resistors prevent damage with 8 ohm
headphones)

If you run a waveform with a DC component through a VCA, the VCA will put an
envelope on the DC also. This will sound like a thump or click. So most
signals intended primarily for audio will be centered around zero volts. For
example +/- 5 volts or 10 volts peak to peak. You can still have a DC
component as a normal part of the waveform. A pulse width modulated signal
has a DC component that varies with the pulse width.

Most signals intended primarily for control, go from zero to 10 volts. This
is because 0 x 0 = 0. An input attenuator (coefficient potentiometer) can
turn them all the way off, without having to reset the manual center point
of the controlled module. Many modules use exponential converters, which
only operate in a positive mode. So negative is really just moving toward
"less" from the manually set center point. You are always somewhere on a
positive curve.

Plus and minus inputs versus plus and minus outputs.

Many systems which use phone jacks, take advantage of the built in switch to
prepatch the modules into some useful configuration. This is intended to
make the instrument more rapidly useable. But it makes it difficult to
disconnect a prepatch. So most systems using phone jacks have a control
voltage scheme where the input attenuators only have a positive coefficient
(0 to +1). This allows you to positively shut off the prepatched signal, but
has the disadvantage of needing a negative going output to get a negative
going control effect. Many of these systems supply patchable voltage
inverters for this reason. In the case of voltage controllable linear
envelope generators, positive and negative outputs allow the straight lines
to be bent inward or outward by feeding a positive or negative output back
to the control input. This allows you to bend linear envelopes to simulate
exponential ones.

Buchla and Serge innovated an input which can have a coefficient of -1 to
+1. This eliminates the need for negative going outputs, but makes it
difficult to rapidly find the 0 point. Most of these systems use banana
connectors, which don't have a built in switch and are easy to just yank
out. So having a positive zero point is not as important. (I should mention
John Blacet's excellent "Mixer Processor 2040" which has Buchla type inputs
with center detent pots and trimmable zero points. This allows you to get a
definite zero rapidly, but at the cost of additional circuit complexity)

Notes:

(1)Including Otto Luening (whom I once had the incredible privilege of
having lunch with) , Pril Smiley and Serge Tcherepnin (who very graciously
spent time on the phone with me answering dumb questions).

(2) In Hertz
1 20-40
2 40-80
3 80-160
4 160-320
5 320-640
6 640-1280
7 1280-2560
8 2560-5120
9 5120-10240
10 10240-20480