Roland has been teasing their new AIRA line (which will include a TR-808 “remix,” the TR-08), and it seems the line will be based around a virtual analog, featuring a technology that Roland is calling “ACB” (Analog Circuit Behavior). This is apparently a very polarizing issue, and lots of people have reacted negatively to the announcement - they think that a computer model of analog technology could never faithfully recreate the original.
I think that is totally incorrect. As part of my PhD studies at CCRMA (Stanford University’s Center for Computer Research in Music and Acoustics), I’ve been researching the circuitry of the 808 for a year, and one of my findings has been that there is no reason that the 808 cannot be simulated with virtual analog.
My research focuses on breaking the circuit into functional blocks, and creating physically-informed analyses and digital models of each block. Even though my research won’t be presented or published in conference proceedings until later this year, I wanted to share something that I believe is a strong counter example to the idea that analog devices cannot be well-simulated in a digital system.
I took the circuitry of the TR-808 bass drum sound generator and entered it manually into LTSpice, a free SPICE (“Simulation Program with Integrated Circuit Emphasis”, technology whose development began at Berkeley in the 70s). Setting the tone knob fully open, and the volume knob at 50%, I ran a transient analysis with a maximum time step of 0.000002267573 seconds (10x oversampled, with respect to CD audio, where fs=44100), exported the output voltage into Matlab, resampled to CD audio with Matlab’s interp1 function, and wrote it out to a .wav file.
The first 12 seconds simulate the case where the decay knob is turned all the way up - the bass drum will ring out for a long time. The next 12 seconds simulate the case where the decay knob is turned down a bit. You can see and hear that the long decay case has all sorts of variation between notes, and the shorter decay case has much less variation. If we turn down the decay enough, there will be hardly any variation between notes.
I think that this clears up one misconception that many hold about analog drum machines, regarding the so-called “machine gun effect.” The “machine gun effect” is the (usually un desirable) effect of playing many identical repeated drum hits. In the case of sample-based or signal-based models, people have created all sorts of workaround to deal with the effect, including randomizing synthesis parameters on each hit, or triggering from a large library of near- but not totally-identical samples.
Although it is usually attributed to the non-linearities of analog circuitry, or some sort of real-world randomness, the dominant reason that TR-808 bass drums (and by extension, other sound generators involving bridged-T networks or envelope generators) can sound different when many notes are played in quick succession is that a filter’s response is going to be added to any response that is already happening (the filters obey superposition). The decaying tail of the previous note and the new note can interfere constructively, destructively, or anywhere in-between. Anyone who has jumped on a trampoline knows the phenomenon of “double bouncing” and the disappointment of a slightly mis-timed double bounce. This is precisely the same concept, and it has everything to do with the fact that it is a filter creating the drum sound, and nothing to do with the fact that that filter is analog.
This simulation actually runs much slower than real-time (it takes about a minute to render a second of audio). One point of my research (besides drawing attention to the ingenious designs of the 808’s creators and supporting the work of drum machine modders and designers) is making models that are physically-informed, but simplified enough to run in real-time.