[sdiy] Analogue hardware simulations in LTspice

rsdio at audiobanshee.com rsdio at audiobanshee.com
Sat Oct 12 23:19:46 CEST 2019

It is an “interesting” idea, but I’ll take a risk and say that it’s rather pointless, at least for linear, time-invariant (LTI) systems. The math behind the Fourier Transform is centuries old, and gives a full model of an LTI system. I suppose that if you’re modeling the distortion characteristics of real analog filters, then some other tool besides the FFT is needed. i.e. distortion is non-linear, so there might be some benefit to fuzzy analysis. On the other hand, even distortion can be modeled directly from the circuits without resorting to buzz words.

I do believe that it was someone like Yamaha who built a robot that could hit each key of a grand piano with 128 different velocities and gathered all of that data for analysis. This was decades ago, before the better digital piano synths were released.

In either case, I think it would help a lot if humans provided some intuitive input to the modeling algorithms, at least if we care about the energy needed to recreate the sounds after the learning process has completed.


p.s. But, most of all: Thanks! to Richie for sharing these LTspice models!

On Oct 11, 2019, at 6:19 AM, Rutger Vlek <rutgervlek at gmail.com> wrote:
> Interesting idea to create an (IR) of hardware without having access to the actual hardware. I recently read that the digital filters of the ASM Hydrasynth were modeled after real analog filters (including Ian Fritz' threeler!) using machine learning on data (test signals) from the original hardware. Doing the same thing through Spice could also be potentially interesting! It allows to bring a non-real-time Spice model into (almost) real-time applications. The question is, how complex a machine learning model would you need to capture the essence of the Spice circuit in sufficient detail.
> Oh and speaking of it... has anyone ever tried doing this with the Yamaha DisKlavier? That could also potentially be used to generate a very nice set of training-data from which ML techniques could learn to predict the signal resulting from a key strike (the input to the model).
> Regards,
> Rutger
> Op do 3 okt. 2019 om 16:26 schreef Ingo Debus <igg.debus at gmail.com>:
>> > Am 03.10.2019 um 15:20 schrieb Richie Burnett <rburnett at richieburnett.co.uk>:
>> > 
>> > For anyone interested in simulating analogue hardware, I've put the LTspice schematics for three bits of classic hardware on my website for download here:
>> Thank you, Richie!
>> Speaking of LTSpice, does anyone know a simple way to get an impulse response of a circuit other than sending a narrow pulse into it? I mean, LTSpice can generate the frequency response of a circuit, so there should be a simpler way, no?
>> Ingo

More information about the Synth-diy mailing list