The circuit schematic was thoroughly analyzed stage-by-stage, including: gain reduction, sidechain and make-up circuit. As it contains about 40 electronic components, a brute force simulation would be too CPU-intensive for real-time use. Fortunately, some simplifications were possible without affecting the overall sound. In particular, we found we could differentiate between actual features of the circuit, and what mostly reflects hardware design constraints of the time, which we could then work around in a virtual analog model. Some stages were decoupled, some components were lumped together, and some components were forwent altogether.
Next we performed measurements on several optical compressors to extract the characteristics of their optical cells, namely the photoresistor’s internal dynamics, and the law for the optical coupling between the photoresistor and the light-emitting element. As these were real machines used in real studios, the measurements had to be as non-intrusive as possible, meaning no circuit dismantling to measure components in isolation. Knowing the rest of the circuit’s electronics already, we could do some reverse engineering and estimate our model’s parameters for the optical cell with a bit of machine learning, based on these measurements.
For the simulation itself, we applied some reduction and pre-resolution techniques, and were able to reduce the model to 2 x 4th order systems (out of 40 components to begin with). The solver was also specially designed to be computationally efficient. Finally, we added modern features including external sidechain, variable responsiveness, frequency response correction, and tube drive, to create our own innovative physical model.