Almost like I made them...
Implementing the various building blocks as analog parts, such as in the earlier synths, is not an ovious task, since the demands on the circuits are that they cover fairly large parameter ranges, are stable (over time and temperature), have good audio quality or attractiv synthesis properties, and preferably would be programmable and yield well-repeatable patches.
All but the latter two demands are in some way fulfilled by the circuits various manufacturers have put out, and it is an interesting phenomenon that various imperfections of relatively old circuits are actually making some of these designs musically attractive. This can be exemplified by the fact that various old analog synthesizers are sold for prices not too far from modern counterparts, and at significant fractions of their original price.
Bob Moog is one of the main innovators in the (at that time mainly analog) sound synthesis area. His synths are famous for their fat and clear sounds, and contained innovative circuitry that can produce sounds that are still very popular, for instance in contemporary house music. At the risk of infringing on some copyrights, I've looked at some of the service book circuit diagrams, and reproduced parts of the original circuitry with parts that should be readily available (and happened to be available in Microsim's Pspice Simulator). They are not exactly the original, but functionally there is hardly a difference for the greater part.
The oscilator is composed of a current source linearly sinking a capacitor's voltage, a schmitt-trigger-ed comparator detecting the zero crossing, and a field effect transistor to shortcircuit the capacitor to its initial discharged state. The feedback circuitry of the schmitt-trigger makes it produce a 300nS discharge pulse which is sufficient to discharge the capacitor and to compensate for the overshoot caused by the charge injected into the fet's gate to make it switch fast (It tranfers hundreds of mAmps peek current). The idea of the original circuit is that the discharge phase should take under 500 micro seconds, so the effect of the (fixed) discharge time on the frequency dependency of the oscilator on the drive current is less than one tenth of a percent for the highest frequency, 20kHz. This circuit is a bit faster, but it does suffer from a touch of overshoot, that is the voltage is not exactly put at 5.00 Volt. Then again, the amount of overshoot is fixed, and I haven't found a spice model for the fet (and comparator) used in the original cicruit.
From the following diagram it can be seen that the circuit oscillates fine at in this case ca 14kHz. This frequency can be determined by the current source at the bottom, which is a part of the original circuit, and contains an exponential convertion, that is, the input voltage is exponentially resonsible for the capacitors charge current. From the charge current, the frequency can easily be derive, considering the Voltage drop of the downward sawtooth is constant and ca. 5.0 volt.
The following diagram shows a blowup of the 'reset' time or dischage time of the capacitor, see time scale and the values from the two vertical markers displayed in the small marker window.
The output of the comparator is a clearly visible pulse, the voltage at the gate is high-bounded when the g-ds voltage is reversed, and the discharge diode is added to make sure the fets gate is drawn sufficiently low very soon after the discharge period is over, otherwise the d-s resitance would make the discharge curve non-linear. I'm not quite sure the lm111 is the best choice for the circuit, since it has an impractical open collector (npn, drive low) output, but it is the only comparator I could easily find for Microsim.
The reason I will like to built this filter, is that it does more than the specified 24 dB / Octave voltage controllable cutoff filtering, it has inherent feedbacks and especially non-linearities that unboubtably make up for its characteric sounds. I will do a dnon-linearity analysis when I find the opportunity, since that is one of the keys to a succesfull digital implementation.
The circuit is quite ingeneously made of four sections of current coupled balanced transistor pairs, which would best be all of the same substrate for temperatur coupling, possibly even with temperature control to stabalize them. The basis-emittor junction is driven with an adjustable current, derived from the cut-off frequency control voltage, to render a varying (virtual) impedance at the emitter. The capacitors are then used to filter out the high frequency components by frequency dependent coupling of the balanced signal paths. Interestingly enough, the emittors virtual impedance can be linearly (the exponential driver is not shown) varied over at least 3 decades, yielding cut-off frequency ranges of e.g. 20Hz to 20kHz. Depending on how much the circuit is modulated with the signal component, the transistors exponential Vbe-->Ic behaviour will show up when the balanced circuit is no longer fed with a 'small signal' exitation. Guitarists now how to appreciate non-linearies, and in this case the symmetrical distortion and the feedback inherent in the emittor coupling in bottom part of the ladder makes the distortion of a non-clipping type.
and again he is forgetting this page is available to the whole world and throws in the Microsim schematic file . The output of the circuit is taken from a simple instrumental amplifier (a difference amplifier), cancelling out common modes stemming from the drive current or unbalance in the circuit, and the amplification of this amp determines the amount of signal fed back to (the other side) of the ladder filters input, thus effectively providing a 'resonance' control, the filter characteristic obtains a peak at the cut-off frequency, which eventually turns the circuit into self-oscilation.
The filter can be successfully simulated (note that microsims package is freely available in demo version, and can directly simulate the above schematics file!), and with a mild amount of resonance yields the following transfer function:
The cutoff-frequency can easily be controlled over the whole range shown by changing the amount of current fed to the circuit from the source at the bottom, and the relation between the cut-off frequency and the current is practically zero-based linear.
Microsim can't do much more circuitry than this in the demo version, so I didn't (yet) put in the original (discrete) intrumentation (difference) amplifier, which no doubt influences the sound of the resulting circuit, and I'm looking at the voltage-current exponential conversion, which in the original version contains a circuit that near zero input voltage (ramp-wise) 'noise-gates' the cut-off frequency to zero. That circuit is bound to add to the dynamical behaviour of the whole circuit.
The circuit diagram below generates a current up to about 1mA which exponentially depends on an input voltage roughly between -10 and +1 Volt.
The graph resulting from plotting the input voltage against a (logarithmically represented) outut current is:
The red line is simply a straight line (target), the green one follows it quite closely over at least 3 decades, which is required to drive the filter. A little tuning is needed to make the relationship exactly 1Volt per octave and in exactly the right range, which can in practive easily be done by pots.
Here is the Microsim schematic file .