
Jim Heintz posted the following on the AH list in response to VST emulation of classic analog. I asked him if it would be ok to put this up here and he have the thumbs up.
"This is an interesting point. Being an engineer and the developer of
TimewARP 2600, I understand very well how emulations (at least the TimewARP 2600) are implemented.
It is a matter of fact that if you want more accuracy in your digital emulation, then more CPU power is required. One simple example of this is by looking at a pulse wave that is generating a 10khz signal on a computer setup with a 44.1khz sample rate. If you set the pulse width of the signal to 10% there is no real accurate way to represent it at the 44.1khz sample rate since there are only approximatly 4.1 samples per cycle of the waveform at that rate(for simplicity, let's not look at oversampling which can defiantly improve the resultant waveform).
The easiest way to improve the accuracy is to increase the sample rate which also has the effect of multiplying the CPU usage by what ever the difference is. If you choose a 96khz sample rate instead of the 44.1khz sample rate, you now have 9.6 samples per cycle, which will give you a much better resultant waveform. This will cost you however by using quite a bit more CPU.
As far as modeling components rather than outputs, the TimewARP 2600 does this to a great extent now. There are certain modules in a synth that cannot be accurately modeled by components however. Oscillators are the biggest culprit here. Because of the fact that you are in a sample based environment, you MUST keep all generated harmonics below the sample frequency otherwise you will hear audible aliasing in the signal. This phenomenon applies to any digital signal, sampled or generated. As such, if you create an oscillator in software that exactly models the circuit (saw, pulse, triangle) you will defiantly get aliasing in the output signal. To produce signals that sound and behave like real analog signals in a digital environment, you must band-limit everything to be below the Nyquist limit (which is the sample frequency divided by 2). On the other hand, many components can be simulated using the components of the circuit.
The best candidate for this type of emulation are filters, however, even in filters, you must be concerned with band-limiting. This is usually done with oversampling in filters.
One of the biggest areas of difference between most emulations an real analog has to do with the rate at which parameters are updated. Many emulations only update parameters and control cv sources at the digital frame rate, which is usually between 50 and 500hz. This of course effects the quality of the output signal. This is most apparent when you have patches that use higher frequency control sources. Audio frequency modulation is out of the question for emulations that use this scheme to update parameters and control cv sources. As a side note, TimewARP 2600 updates ALL parameters and ALL sources at the full sample frequency.
To the point that no two vintage analog devices sound the same, this is quite true. Alan R. Pearlman told me that when they were building synths they spent a great deal of there resources matching an qualifying components so the circuits would behave as consistently as possibly. There was always variations that they could not control. When you implement an emulation, there is inconsistencies go away and the resultant output is always the same for a given patch. In order to introduce "life" into the emulation, we do add stochastic behavior in certain places, otherwise the emulation would not feel right. It would not be out of the question to add features to an emulation to allow the user to adjust certain component values and thereby hear the difference, but we (at least not Way Out Ware) have not produced that product yet. If there is enough demand for it, then I would love to take a project like that on.
Best regards,
Jim Heintz
Followed by:
We have had many reviews of the
TimewARP 2600, and most have been overwhelmingly positive. As it turns out, the people at
SOS reviewed the first release of TimewARP 2600, and spotted several bugs that we too spotted and fixed in the updated version. It was unfortunate that they did not update the article to match the latest version of the program since the article was printed about 9 months after the update was made available. Also, I believe their ARP 2600 was from a different era than the one we modeled. That said, I have contacted SOS, and they have told me that they would review the next update version of TimewARP 2600 when we make it available. We hope to have it out soon.
For another point of view, you should check out the EM article written by Larry the O. Larry at one time had a very large collection of ARP hardware and is an expert on the ARP 2600. He found the TimewARP 2600 to be quite convincing.
We are about to release our second product which is called KikAXXE. It is an emulation of an ARP AXXE synth with a step sequencer, drum machine (ala TR-606) and a tape delay similar to an Echoplex EP-2 built in. It is designed to be easy to lay down beats and synth lines. It takes advantage of features that are present in the computer such as host sync, midi learning of beats and synth lines, etc, as well as calculating an analog style signal path for all of the audio to achieve excellent audio quality. It uses the same core engine as TimewARP 2600. We will make more info available after it ships. We also plan to have a demo available on our website similar to the demo for TimewARP 2600.
Best regards,
Jim Heintz"