New Stuff

[Haible Juergen]

	>IMHO the main reason is that they are required to have an extremely
high
	>gain in an open loop. In a VCA there is no global feedback which
would 
	>potentially reduce the gain, moreover we usually operate on
somewhat stan-
	>dardized line levels (at least in modulars it should be so),
therefore we
	>do hardly need any gain, rather a controllable attenuation.

The intrinsic function of a differential pair is that of a variable
transconductance,
and as someithing like "100uA is larger than 5mV" doesn't make much sense,
it's up to us (by choosing the circuit environment), to make that a voltage
gain
or voltage attenuation. You're completely right, the general case is that
the
integration capacitor in most VCOs has a higher voltage swing than the final
synth output amplitude that goes to the mixer. So there is an overall
attenuation
rather than amplification.
The strange thing is, even if we build something like a "Voltage Controlled
Attenuator"
from a 100k / 200R divider, CA3080, 100k load resistor, unity gain buffer,
we
do not really use the 3080 as an attenuator. With the 100k load resistor,
the 3080
will amplify a few millivolts at its input to a few volts at its output. It
*must*
act as VC amplifier, because you did some fixed attenuation beforehand !
It's clear that attenuation to the 20mV range must happen at least once in 
a synth (unless you're using gilbert multipliers, and even there ...), but
there is 
no need to do this on cascaded VCA stages, or from the VCF to the VCA. 
Not even for the resonance feedback loop: look at 
http://www.synthfool.com/diy/jh2040.gif 
and compare it with the SSM2040 application: a simple pnp transistor pair is

enough to replace a 20dB voltage amp and a VCA.

	> Am I right to suspect more stages mean higher dynamic range ?

Why ?

Without VCAs, more gain stages and a tight feedback loop means a slightly 
higher dynamic range for, say,  < 0.1% THD, and you buy this with sharp
corners of your transfer function outside this range. OTOH the dynamic
range for, say, < 5% THD limit  is not notably increased at all. So with
signals
where 5% THD don't hurt, like a combination of a few correlated VCO waves,
you don't get any notable difference in dynamic range, but the great
advantage
of soft clipping when you exceed this range, in case you have chosen a low
number of stages / low feedback solution.
That was even without VCAs !

Now *with* differential pair or 3080-like VCAs, you don't have a choice
against
rounding of the transfer courve at all (other than compromising with noise).
So why combine them with high gain / feedback circuits - they just introduce
the effect of harsh clipping, while offering no remedy for the nonlinear VCA
courve.

Things are different when other types of VCAs are used, namely gilbert
amplifiers,
or more modern variations. These make the argument of the last paragraph
irrelevant,
and bring you back to the choice of the last but one paragraph ("without
VCAs").
Even there, my choice is the slightly nonlinear one in most (not all) cases.

Things are completely different for complex signals (the "HiFi" case), of
course.


	>> To answer your question: In the Minimoog, one stage was used for
VCA
	>> envelope,
	>> one stage was used for Pedal Volume control, and the last stage
is a class
	>> A buffer amplifier / line driver.
	>
	>Well, sasn't it easier to control the first stage by a sum of ADSR
CV and 
	>the external input? 

No. The VCAs are basically linear, and you want a multiplication of the ADS
VC
and the pedal, rather than addition. You could use an exponential VCA
instead,
run the ADS and Pedal thru a log function, and feed the sum into the (now:
single)
expo VCA.

	>Finally, what do you think about differential pair with fully
balaned dynamic
	>load by a current mirror ? Like the one proposed by Roman at
> http://www.multimania.com/romcio/m_modvca.gif
	>I must admit I love the simplicity of his ciruit.

I love it, too - the signal path is basically that of the gm cells of the
SSM2040
VCF and SSM2024 VCA.
Note that you buy the omission of 2 current mirrors (compared with the 3080
scheme)
with restrictions in output voltage swing. The Moog VCA goes even one step
further: More restrictions on the output levels (needs symmetrical input of
next stage),
and one more current mirror spared.

BTW - I have slight questions about the biasing of Roman's circuit,
especially
about the dc voltage levels around T2. Does it really work with the bases of
T2
and T3 at approximately the same level ? Roman ?

JH.