RMS stabilization of oscillators

Following on from the stablization of sine wave oscillators, the same problem applies to more complex state variable type oscillators, such as the primary VCO core of the Oscilloplasm. This is based around the classic state variable circuit, though with some added complexities. Here is a photo using classic analogue computing notation. Just as an aside, when this circuit was first envisioned, opamps were realised with vacuum tubes, and the servo-multipliers were electromechanical devices that relied on physical rotation to generate the desired result. This technology was used in early rockets and aerospace machines as well as the laboratory. Don’t knock the humble transistor!


T is effectively a saw wave that frequency modulates one side of the sine core, the potentiometer 2 adjusts the time constant of the other side to vary the frequency content of the output.

Zener diode negative feedback and diode clipped positive around integrators 3 and 4 is what was used in the first iterations of the module, but the sound was not pleasing. Simply allowing the VCO to distort by clipping the rails actually results in an interesting sound and this was genuinely considered, the problem with this however was not the distortion but the amplitude instability, as frequency stability depends on this, and this is essential for a high performance modern oscillator.

The end of the journey was found in the THAT4301 RMS to DC converter chip. RMS to DC conversion ICs typically have no provision to vary conversion rate, which really rules them out for use in an oscillator core, because successful RMS to DC conversion has only a small useful operating region where conversion time is low enough to remove all the AC components of the signal, but high enough to respond quickly and avoid amplitude modulation from long settling times.

The because the filter in the 4301 works in the log domain, integration time can be varied via a bias current into pin 2. Rather than using a fixed bias as suggested in the datasheet, we can use a variable one, like the same value that is controlling frequency in our oscillator or filter.

The principle is the same as other AGC (automatic gain control) techniques. We use some component to detect the amplitude level, in this instance averaged over several cycles, relative to some reference level, and apply this to a gain control element that allows for proportional negative or positive feedback around a specific node in the VCO loop (in this instance an integrator in the VCO core).