Modern virtual analog modeling synthesizers pretend to emulate former old school analog synthesizers up to 100% but not all of them seem to have taken common problems and limitations into account the analog synthesis "suffers" from. Here you will find some typical problems of the "incomplete" VAM technique. Well one has to note that these problems are in fact existing, because the synthesis is "too complete", meaning the model generates a clean and most accurate sound which is considered to be "dead".

Regarding the classical SST-waveforms (sine, square, triangle) there are several options to generate them:

	Analog Synthesis	Clean Digital Synthesis	Common VAM-Synthesizer	96kHz.org VAM-System
Sine Wave	self oscillating filters, free running	strict DDS with increasing phase vector and lookup table for the amplitude introducing digital phase jitter and amplitude noise	virtual self oscillating filters with shaped amplitude and predefined phase, independent from each other	virtual self oscillating noisy non linear filter based on energy loss model and natural damping were phase and amplitude are the result of acceleration process when fed with energy
Square Wave	Schmitt triggered HILO amplifier with noise dependent phase derived from sine wave	50% decider HILO derived from phase vector	50% decider HILO derived from phase vector with artificially added dirt effects and band with limitation	noisy Schmitt-trigger model with smoothed switch point forwarding artifacts from the sine wave generator leading to natural phase errors, instable frequency and band limiting
Triangle Ramp	integrator ramp performed by operational amplifier fed with square signal from above and turn point circuit for triangle	50% decider UPDOWN derived from phase vector	mathematical integration (up down summation) of the ON-OFF signal from above with added band limitation and slight amplitude offset which has to be corrected	noisy auto switching up-down integrator circuit with non linear amplification smoothed switching and natural dynamic DC-offset caused by square stage leading to dynamic band limitation and frequency instability

The clean way is easy to implement at first sight but leads to a boring sound in the first place. Furthermore it can introduce some unwanted errors. See the article sound generation with DDS for some details. Regarding calculation power and resource usage, different way appear more appropriate for this:

Virtual Electrical Synthesizer Model for SST

This is the first electrical model from 1993 generated and simulated in the MicroSIM pSPICE analog simulator on a Vobis DX33 PC.

Later the model was replaced by mathematical equations and optimized using the C-language based compiled models for the MicroSIM Design Center together with Visual Studio 4. The usage of C-language made it easy to move the model to a real DSP later. A first approach was done in 1996 with a TMS320-System from Texas Instruments comparing VA-VCOs and NCOs as well as some filtering. Also a mathematical model was used in the ABM-blocks with Micro SIM Cadence. These blocks offer analog behavioral model calculation. The next step was the implementation of the algorithm into the DSP 56301 based evaluation system from Motorola in 1998. This DSP was strong enough to create a larger number of self oscillating filters and also additional modules to create a fully working synthesizer with. Later this was moved to the Chameleon Sound System.

Sound Samples:

Sinus Clean

Sinus VAM96

Triangle Clean

Triangle VAM96 

Read More:

Analog Modeling with DSPs and FPGAs