Jörgen's DIY VCA shootout

A good VCA is important for the sound quality in an analog synthesizer. There are a number of special purpose VCA circuits that satisfy these needs. The problem with most of them is that they are either expensive, hard to find, out of production or all of the above. I wanted to find out which is the best VCA that you can build with cheap, easy to get parts. To test the circuits I built prototypes on Veroboard. The circuits were adapted, if necessary, to give unity gain at 10 V CV and accept an input signal of 10 V peak to peak with some headroom.
The measurements were done with an Audio Precision System Two. Noise level was plotted at different frequecies by sweeping a bandpass filter in the AP. This was done with the VCA control voltage at both 0 and 10 volt and the input grounded. Noise was also measured with the signal generator in the AP swept together with the bandpass filter. This shows the signal bleedthrough at different frequencies. In this measurement the signal level from the AP was 10 volts peak to peak. Of course the control voltage in this case was 0 V.
Signal to noise was also measured without the bandpass filter but with an A-weighting filter. This was done with no input signal and with three different input signal frequencies (1, 2 and 10 kHz). Without signal, measurements were made with the VCA closed (0 V CV) and opened (10 V CV). To check the distortion at different signal levels, the singal amplitude was swept between -60 and +20 dBv and the THD+noise was plotted. This also showed the lowest possible distortion and the headroom. The latter was considered to be the level where the distortion exceeded one percent. If this level was lower than 10 V p-p (10,968 dBv) the circuit generally was changed and new measurements were made.
Frequency response was measured by sweeping the generator in the AP. All circuits had practically flat response throughout the audible range.
Distortion was also measured by sweeping the frequency at two different input levels (0 dBv and 11 dBv). It turned out that the distortion did not change noticeably with frequency. Therefore the plots are not included.

Measurement summary